Sequences

Guide version 1.62.0 · For Sequences 0.116.0

User guide

How to search patent and public sequence sets, read the results by patent, by applicant and by alignment, and keep a record of each search.

Start a search

Sequences runs in three ways:

  • The hosted service, at sequences.conductscience.com. Sign in with your ConductScience account.
  • A server your organization runs, the server edition. Sign in with your organization's account.
  • The local workbench: Sequences with sign-in off, on one computer, for one person. It needs no sign-in. See Sign-in off.

Open the search page, then:

  1. Paste one or more sequences into Query sequences, or choose Open a file. Query type reads DNA / RNA or Protein from the letters; choose the other one if the reading is wrong.
  2. Under Search in, check one or more sets, such as Patent proteins, and choose a search method. Two or more sets run as one combined search.
  3. Give the search a Search name if you like, then choose Run search.

To try it, choose Use demo data beside Open a file. For DNA / RNA it fills the primers and probe of US12788762B2 (us12788762b2-primers.fasta); for Protein, the 16 peptides of US12788009B2 (us12788009b2-peptides.fasta, and the same peptides as PIR records: us12788009b2-peptides.pir). See Demo data.

Search in lists the sets of the query's type first. Translated search, under them, lists the sets of the other type, unchecked; a search there reads the DNA in all six frames.

Query typeSets of its own typeSets under Translated search
DNA / RNADNA sets, by blastnProtein sets, by blastx: the query translated in six frames
ProteinProtein sets, by blastpDNA sets, by tblastn: each DNA sequence translated in six frames

To search an antibody's CDRs, turn on Antibody CDRs beside the search card's title. The card then shows the antibody form, described under Antibody CDR search below.

The query box takes FASTA, with a >name line before each sequence, PIR records, or plain letters, which become one query named "query". It takes up to 20 queries, of up to 20,000 letters each and 100,000 in all. Spaces and digits are left out, and in RNA a U is searched as T. Reverse complement replaces each DNA query in the box with its reverse complement. Where no suggested search can run, Try an example query fills the box with a made-up DNA query.

For more queries, up to 2,000 and 2,000,000 letters in all, choose Save as a query group under the box. It shows when the box holds more than one search takes. The queries become a group under Query tools, and the group's search opens at the set and method the form has chosen. The group runs as one batch, up to 20 queries in each search, so Hit by several queries can count it. Save as a query group beside a listing's Start the searches saves its picked sequences the same way.

A PIR (NBRF) record starts with a header such as >P1;CRAB_ANAPL, which names the sequence after the semicolon. The next line describes the sequence, and may be blank. The sequence follows and ends with *. The header's two letters say what the sequence is: P1 a protein and F1 a protein fragment; DL and DC linear and circular DNA; RL and RC linear and circular RNA; N1 and N3 other RNA, N3 for transfer RNA. With XX the letters decide. A file may hold several records, and FASTA and PIR records can share a query.

Each search runs as a job. Recent jobs lists your searches: open a finished one to see its results, or cancel one that is still running. A search that failed, was cancelled or was interrupted offers Run again on its card. Find a search keeps the jobs whose name, or set, holds your words, and the status choice keeps those waiting or running, done, or failed or stopped.

Query type

The page reads the query's type as you paste. Letters read as DNA or RNA when every one is a nucleotide code and at least 90% are A, C, G, T, U or N; anything else reads as protein. The line beside the chips says where the type came from. A click on a chip sets the type until you empty the box.

  • A short protein made only of A, C, G and T reads as DNA. Choose Protein.
  • A PIR record's header gives its type, and the line beside the chips says so. A chip that says otherwise is refused under the box.
  • RNA reads as DNA / RNA, and its U is searched as T.
  • A query that holds both DNA and protein records is refused under the box. Search one kind at a time.
  • For a pattern search, the query type stays as you chose it.

A fault in the query shows under the box, names the line and the character, and selects it in the box. The message in the corner is for faults of the service.

Drop a file

Drop a file on the search card, or choose Open a file, and it goes where it belongs:

  • A FASTA or PIR file fills the query box.
  • A sequence listing, in WIPO ST.26 XML or ST.25 text, is imported under Sequence tools and opens its SEQ ID NO picker.
  • A publication's full-text XML with claims offers Start a case from the publication's number. The case takes its claims, and its listing too when the file holds one. When the listing was filed as a file of its own, the offer says so; import it under Sequence tools.

To try it, drop us20260345831a1-with-listing.xml, a published application whose XML holds its listing.

While a search runs

The status bar under the search card follows your search. The tab's title puts its state before its name: "Waiting:" or "Running:" until it ends, then "Done:" until its result opens, or "Done in part:", "Failed:" or "Interrupted:". A finished search opens its result by itself, unless you opened another result since it started or you are typing in a field; then it waits behind Open results. After a reload, the status bar follows your newest search that is waiting or running.

A search's state reads the same on its card, in the status bar and in the tab's title: waiting, running, done, failed, cancelled or interrupted, and done in part for a combined search whose sets did not all complete. Choosing Run search again while the same search waits or runs starts no second search: the page says "This search is already waiting." (or running) and follows the one under way. While the service does not answer, as during an install, the status pill says "Not answering. Trying again." and the page keeps asking until it answers. A search the server no longer has reads "This search is no longer on the server.", with Close.

A running search says how long it has left, in whole minutes, from the server's estimate. A Smith-Waterman or global search of a DNA set also counts its queries as they finish, as in "2 of 5 queries done, about 6 minutes left", and works out the time left from their pace. When a search runs past that time, the line leaves the time out.

Your organization may run Sequences on several servers that share one queue. A search of a snapshot then runs on whichever server is free, and its result is the same wherever it ran. If that server stops, the search reads "waiting" again, keeps its place and starts over on the next free server. A search cut off three times this way fails with "The search was cut off 3 times before it finished. Submit it again." When the server you sign in to restarts, a search of a snapshot that waits keeps its place.

Searches that ask first

Your organization can have Sequences check with you before a long search. A search of a large set, such as TrEMBL, or one whose estimate passes the time your server sets, then asks before it queues: "TrEMBL is a large set. This search should take about 12 minutes. Run it?" Choose OK to run it, or Cancel to leave it unstarted. Searches you start together, such as a search of several sets or of a listing's sequences, ask once. A program sends such a search again with "confirmed": true. See Programs and API tokens.

Your organization can also set how much processor time each account's searches may use in a day. Once yours have used it, a new search stops with a message that names the limit and when you can search again, at 00:00 UTC. Searches already waiting or running finish.

Addresses

Each result has a page of its own: /results/<id> for a search and /results/group/<id> for a combined search. Bookmark it to come back to the result. The page shows the result alone, with the search's name in the browser tab. New search, at the top, goes to the search page, which keeps the form, Recent jobs and the suggested searches. A search that still waits or runs shows its progress on its page, and the page becomes the result when the search finishes. Older links, /?job=<id> and /?group=<id>, open the same page, and so does the results link in a watch email that has one. If you are signed out, the page asks you to sign in, then opens the result.

A case at one of its steps has /?case=<id>&step=<step>. Back goes to the page before, and Copy link, in the result's header, copies the result's address. An address opens only for the account that ran the search; for anyone else, and for a search no longer on the server, the page says "This search is no longer on the server." Back or Forward to a search no longer on the server closes the result that was open. If you open one result and then another before the first shows, the one you opened last shows. A case no longer on the server, whether from its address, from Back or Forward, or from its card in a list read before it was deleted, shows the list of cases, and the page says "This case is not on the server."

An address the server does not know opens a page that says so, with Go to the search page and links to these docs and to All pages. All pages, also in the footer of every page, lists every page the server has: sign-in, the search page, API tokens, the plans page on the hosted service, these docs and the sample results.

Follow-up searches

When a search finishes, Follow-up searches, over its list, suggests searches to run next. Fixed rules read the result's own numbers, and no language model takes part. Each suggestion names its rule, says why from the result, such as "1 hit, under 10.", and names the search it would run, such as Search Swiss-Prot. The rules go in this order, and the first three suggestions that apply show:

RuleWhen it appliesWhat it suggests
Few matchesFewer than 10 hits, or none as close as the identity you asked forA more sensitive method on the same sets, then the same method on a set the search did not read
Hit limit reachedThe search kept as many hits as its limit allowsThe same search keeping up to 5,000 hits. Where it kept that many already, the same search narrower: before the case's prior art date, at a higher identity or at a higher coverage, the first that applies
Part of the query unmatchedThe best match covers under 80% of the queryThe part it leaves out, searched as a fragment
Other molecule typeA protein search read no DNA set, or a DNA search read no protein setThe same query, translated
Antibody chainThe query reads as an antibody chain, on a server where antibody numbering runsA search of its CDRs

After BLAST, the more sensitive method is Smith-Waterman for a protein where FASTA36 is installed, and otherwise Exact text for a short query (DNA under 50 bases, protein of 30 letters or fewer), which reads every stored sequence for its letters. After Exact text, it is Smith-Waterman or BLAST. A set offered on request, such as TrEMBL, comes once the search has read every set offered now.

  • Run fills the search form as you would and starts the search. Open the new search, beside the suggestion, opens its result.
  • Apply filter puts a higher identity or coverage on the list you are looking at, with no new search.
  • Dismiss hides a suggestion, and the next one that applies takes its place.
  • The Search record tab lists each suggestion shown and what became of it: Shown, Run, Applied or Dismissed. A search started from a suggestion names the search it follows, and Open the search it follows opens it.

In a case, the results of a search the plan ran show Add to plan in place of Run, and leave out the translated and CDR searches, since a plan compares like with like. Add to plan puts the search in the draft plan with the rule as its reason. A locked plan unlocks and takes it in its next version. Under each claim feature, References suggests searches too: see The References step. The case log records each suggestion shown, added or dismissed, with its rule. Each later version of the plan runs its searches again, and what became of a suggestion holds there: one added shows the plan version and search that hold it, and one dismissed stays hidden.

No match

A result with no match says "No match." over its follow-up searches. Few matches applies to every search that finds nothing, so it suggests a more sensitive method or another set. With nothing to suggest, the result says to try another search method or another set in the search form. See Follow-up searches.

Suggested searches

On a server with a snapshot of the patent sets, Suggested searches, over the search card, offers real public sequences to try, each linked to the record it comes from: the GLP-1 peptide against Patent proteins, the coding sequence of human insulin against Patent DNA, green fluorescent protein against Patent DNA read as protein, the antibody trastuzumab by its six CDRs (where antibody numbering runs), and the peptide SIINFEKL as exact text. The panel is open until your first search, then folds to one line. Try a suggested search, at the top of the page, opens it.

Run fills the form as you would and starts the search, named "Suggested:" and the card's name. When it finishes, its result opens, unless you have opened another result or are typing in a field. Edit first fills the form and leaves the search to you. The result of a suggested search shows Things to try over its list: two or three steps that use it. Close hides them for that search in this browser. A suggested search is an ordinary search and stays in Recent jobs.

Examples and sample results

The examples page, /samples, shows the main parts of Sequences working, read-only, before you sign in. It has a section for search, results, tools, claims and plans, model features (Pilot) and benchmarks, each linking a sample result, a part of the sample case or the benchmarks. When a server has examples, its sign-in page says "Not ready to sign in? See a sample result.", and the menu at the top of every page has Examples.

A sample result is a search of a real public sequence, saved and shown read-only, so you can see what Sequences finds before you sign in. A sample's page is laid out as a result's page (see Results), read-only: its name and what it searched at the top, Copy link and Download as Excel in its header, then the search's facts, the charts and the Patents tab. Copy link copies the address of the view you chose. A sample lists its patents as the Patents tab does: by family, by publication, or under a heading for each applicant, patent office or year, sorted by Query Match, identity, date or sequences found. Set Published before to fold the later publications, as in Dates. Columns chooses the measures each row shows and their order (see Columns), and Download as Excel saves the sample's tabs and its search record in one workbook (see Exports). Each row opens to its best alignment and, in a family, its members. Above its patents, a sample draws the three charts of its hits, with their tables and downloads. A sample cannot be marked or run again: sign in to run your own searches. How to read a result walks through one hit of the GLP-1 sample, letter by letter.

The sample case is a real published patent made into a case on an install of Sequences, searched and saved read-only. Claims and summaries shows its claims with the features read from them, the listing check and the summaries, with the case's report as PDF and Word; each model draft (Pilot) of a summary has a link of its own. Plan and results shows its search plan with each search's reason, then the results and the references. Claim suggestions (Pilot) shows the saved record of a model's claim suggestions on another published grant. Each suggestion has a link of its own and shows its quote, the checks that kept it for review or dropped it, in the order the service runs them, and what made it: a language model, named by its role, and the first characters of the prompt's SHA-256, which the record above the list gives whole. Nothing on these pages can be marked or run.

When your sign-in ends

A sign-in lasts a set time: 7 days on the hosted service, and on a server your organization runs, the time its staff chose (12 hours unless they chose another). Five minutes before the end, a message over the page says when. Choose Stay signed in to sign in again in a new tab. Once you are signed in, that tab closes, the message closes and the page says until when. Your work on the page stays as it is, such as a query you typed or an open result. Not now closes the message.

If the time runs out, the message opens again to say your sign-in has ended, with Sign in again, which works the same way. A search that was running keeps running. API tokens are not sign-ins: each lasts as long as it was made to. While it has ended, the page checks on nothing: not Recent jobs, a search that runs, a listing that imports or a watch that runs, nor an open case's runs, an import or a model's answer. A case's steps say "Your sign-in has ended. Sign in again to continue." Once you sign in again, they go on by themselves.

Demo data

Each input on the search page, and the name on the API tokens page, has Use demo data beside it. It fills the input in one step, as if you had typed it or chosen the file, and nothing runs until you run it. The demo files are small and made only from published patent documents and public database entries: the peptides, claims and listing of US12788009B2, the primers and probe of US12788762B2, trastuzumab's chains from PDB 1N8Z, and the coding sequence of human insulin. Each section of this guide names the demo files it uses.

Each file also downloads at /demo/ and its name, open without sign-in, so you can save it and upload it yourself. README.md lists every file with its source and SHA-256.

A case started with Use demo data under Start a case is a demo case while it holds only demo files: the demo claims, the demo listing and the demo result files. Its card and its header say Demo case, and its header has Delete this demo case, which asks first, then deletes the case with its plan's searches and imported result files. Searches you added to it by hand, and its listing, stay. A file of your own in it makes it an ordinary case, which stays until an admin deletes it.

Sets

A snapshot is a read-only copy of public sequence files, built on a known date. Each sequence in it is stored once and linked to every record and publication that holds it. A set, such as Patent DNA, is one collection in a snapshot. The page lists each set from the newest snapshot that holds it, with how many sequences it holds.

SetWhat it holdsSource
Patent DNADNA and RNA from patent publications, US published applications among themThe patent division of GenBank, from NCBI, and the US patent office's weekly full-text files and sequence listings
Patent proteinsProteins from patent publications, US published applications among themThe patent protein files of EMBL-EBI, from the US, European, Japanese and Korean patent offices, and the US patent office's weekly full-text files and sequence listings
Swiss-ProtReviewed protein entriesUniProtKB/Swiss-Prot
TrEMBLUnreviewed protein entriesUniProtKB/TrEMBL
ESTsExpressed sequence tagsThe European Nucleotide Archive (ENA)
Drug proteinsThe protein chains of drugs, one record per chain: antibodies, other proteins and peptidesChEMBL, from EMBL-EBI, under CC BY-SA 3.0, and the public data of GSRS, the substance registry of the FDA and NCATS
Drug nucleic acidsThe strands of oligonucleotide drugs, one record per strand, written 5' to 3'The public data of GSRS
  • A new search starts with every set of its type checked that the page may choose for you: Patent proteins and Swiss-Prot for a protein, Patent DNA and ESTs for DNA. Uncheck the ones you do not want.
  • The page never selects TrEMBL for you. When you choose it, a line under the sets says how long the search should take, such as "TrEMBL is large: this search should take about 5 minutes."
  • Drug proteins and Drug nucleic acids show where the server holds the drug chains, after the other sets. The page never selects them for you either, because their records carry their sources' own terms. Each record is one chain of a drug, named by the drug, the chain's kind and the source record it comes from, and links to the drug's ChEMBL compound page or FDA UNII record. The records give no date, so case plans do not search them.
  • The hosted service also offers NCBI core_nt, a large DNA database outside the snapshot, which the page never selects for you either. It runs BLAST only, so the choice of method does not show for it.
  • The patent sets hold US published applications, from the first, published 15 March 2001, and the US patents granted since 2001, from the US patent office's weekly full-text files and the sequence listings it publishes. An application and the patent granted from it share one sequence line when their letters match, so one hit lists both. Each week's applications and grants arrive as an addition: see Collection below.
  • Sources and terms, below the sets, names the source and release of each set, its terms and its notices. Where the server has them, it also names the sources of paper citations and marks and of applicant suggestions.

A snapshot can get additions: newer records added after it was built. When a set's snapshot has them, choose under Collection: Main plus New, the snapshot with every addition, which is chosen first, or Main, the snapshot as built. A search through the API reads Main unless its request names the additions.

Only the local workbench can import a FASTA or PIR collection of your own, under Import a database: a file of up to 20 MiB and 100,000 sequences. A search of it runs BLAST, with up to 500 hits for each query (100 unless you change it under Search settings). The hosted service and other servers with sign-in do not take imports. Use demo data fills the form with s1-patent-proteins.fasta: the 26 proteins in the listings of three published grants.

Search several sets at once

Check two or more sets under Search in to run one combined search: up to 8 sets, under one name, with one status bar and one result. A collection you imported is searched on its own. To try it, run the demo query of Start a search with two or more sets checked.

  • Translated search lists the sets of the other type than the query. There a protein query runs tblastn on a DNA set, reading each DNA sequence in all six frames, and a DNA query runs blastx on a protein set. These sets start unchecked. Exact text, Pattern, Smith-Waterman and the global methods compare like with like, so they leave these sets out.
  • The status bar names each set as it goes, such as "Patent proteins done. Swiss-Prot running, about 2 minutes." The time counts how many searches the server runs at once, with any servers that share its queue. Open results so far opens what has finished. A set still searching adds its references to the list when it finishes, and the row you selected stays selected.
  • The result opens on References, one list for every set. A row is a patent publication or family, or a database entry such as a Swiss-Prot accession. It shows the best Query Match and identity across the sets and queries, the earliest date, and a chip under Found in for each set that found it. A patent that Patent proteins found, and Patent DNA found translated, is one row with both chips.
  • Sort, Group, the date fold, the filters and the review desk work as in By patent. The Found in filter keeps what one set found, and Kind keeps patents or database entries.
  • The pane lists the reference's matches by set and query, best first, under the best match's alignment. Choose another match to see its alignment.
  • Keep and Exclude mark the reference in each search that found it, so the mark shows in each set's own result too.
  • Applicants shows when the combined search holds one patent search. Search record lists each set's search with the SHA-256 of its own record and a button that opens that search by itself.
  • Export matches (CSV) saves one line per reference that passes the filters, with the sets that found it. Download search record saves the combined search: its sets, each set's search and the SHA-256 of its record.
  • If a set does not complete, the combined search reads done in part, and the line over the list names each set that did not complete, a cancelled one too, with a button that runs that set again inside the same combined search. Run again runs every set again as a new combined search.
  • Watch this search stays on each set's own search. Recent jobs shows a combined search as one card that names its sets.

Search methods

A search of a set in the snapshot offers these methods. A database outside the snapshot, such as core_nt, runs BLAST.

MethodWhat it doesRuns on
AutomaticPicks the method by the query, and the search record states the pick. See The automatic choiceAll
Alignment (BLAST)NCBI BLAST+, with up to 5,000 sequences for each query and 5 alignments for each sequenceAll
Exact textFinds the query letter for letter in every stored sequenceDNA to DNA, protein to protein
PatternFinds a pattern in every stored sequence. See Exact and pattern searchDNA to DNA, protein to protein
Within k changesLists every stored sequence within k changes of the query, with no score or E-value cut. See Within k changesDNA to DNA, protein to protein
Smith-WatermanA full local alignment with every sequence in the set, by FASTA36 ssearch36DNA to DNA, protein to protein
Global on the queryAligns the whole query with the part of each sequence it fits, by FASTA36 glsearch36DNA to DNA, protein to protein
Global on both sequencesAligns the whole query with the whole of each sequence, by FASTA36 ggsearch36DNA to DNA, protein to protein

The three FASTA36 methods show only where FASTA36 is installed. The Sequences container image, which the hosted service and the server edition run, includes it. These methods return up to 5,000 sequences for each query, each with its best alignment. On DNA they search both strands, and a sequence can return once for each strand.

On DNA, Smith-Waterman finds short matches that BLAST misses. In our benchmark on Patent DNA, run when it held 28.8 million sequences, Smith-Waterman found 16 of 20 DNA fragments of 21 to 50 bases at 80% identity, where BLAST found 8. Each query reads every stored sequence, so a DNA search takes minutes, and longer on a larger set. In the benchmark, a Smith-Waterman query took about a minute at 21 bases and 5 minutes at 308 bases, global on the query took about 70 seconds up to 60 bases and 4 minutes at 308, and global on both sequences took about 25 seconds at any length. Patent DNA now holds 76.2 million sequences, with US published applications, and on the hosted service in November 2026 a Smith-Waterman query of 21 bases took 69 seconds. A DNA search whose estimate is longer than the server allows for one search, 60 minutes unless your organization set another limit, does not start, and the page says so with the estimate. Search fewer or shorter queries at a time, or use BLAST.

BLAST keeps hits with an E-value up to 10, or up to 1,000 for short queries: DNA under 50 bases and protein of 30 residues or fewer, which also run BLAST's short-query settings. The FASTA36 methods keep hits up to 1,000 for protein of 100 residues or fewer and for DNA under 50 bases, and up to 10 above that. Search settings has a Maximum E-value and a Maximum hits per query, and a value you set there applies to every query of the search.

BLAST, exact text, pattern, Smith-Waterman and both global searches return the same hits each time you run the same search on the same snapshot. FASTA36 fits the E-values of Smith-Waterman and global searches to a sample of up to 60,000 scores, and Sequences runs a FASTA36 build that draws that sample the same way on every run. So these searches return the same sequences, scores and E-values each time, whatever the server's load or number of threads. The scores are FASTA36's own. The E-values can differ from those an earlier release of Sequences gave for the same search, so a sequence near the E-value cutoff can appear in a result that lacked it before, or drop out of one that had it.

Patent sequences often write X (Xaa) where a residue is left open. Smith-Waterman and global searches score an X in a stored protein as 0 against any residue, where it used to count as a mismatch, so a stored sequence written with X at a few positions ranks higher than before and is less likely to drop out of the list. An X still scores less than the residue it stands for would, so a sequence with X can still fall past the E-value cutoff of a short query, such as one of 8 to 12 residues, or below the 5,000 sequences a query keeps. A positive score for X would fill the list with sequences that are mostly X. Query Match scores X with the search's matrix, as it does for BLAST.

Choosing a search method says which method fits which query, with recorded runs of each.

Scoring settings

Choose Alignment (BLAST) or a FASTA36 method, and Search settings shows that program's settings, each with its default and its range. A setting left at Default runs with the program's default. With Automatic, the panel says "Choose a method to change its settings.", and Automatic keeps its own choices. Exact text and Pattern have no scoring settings, and Within k changes takes only k and what counts as a change. A search of protein sets and DNA sets together takes none either: search sets of one type to change them.

SettingProgramsDefaultValuesUse demo data
Scoring matrixblastp, blastx, tblastn, and the FASTA36 methods on proteinBLOSUM62. BLAST gives protein of 30 residues or fewer PAM30BLOSUM45, BLOSUM50, BLOSUM62, BLOSUM80, BLOSUM90, PAM30, PAM70 and PAM250PAM30
Gap costs, to open and to extendEvery BLAST and FASTA36 searchThe pair BLAST uses with the matrix, such as 11/1 with BLOSUM62 and 9/1 with PAM30. DNA: 5/2 in BLAST, 12/4 in FASTA36BLAST: the pairs BLAST takes with the matrix or with its DNA scores, which the form lists. FASTA36: 0 to 99 to open and 1 to 99 to extend9/1, the default with PAM30
Word sizeBLAST3 for protein and translated searches, 2 for protein of 30 residues or fewer. 11 for DNA, 7 for DNA under 50 bases2 to 7 for protein and translated searches, 4 to 100 for DNA2
Low-complexity filterBLASTOn for blastn, blastx and tblastn. Off for blastp and for DNA under 50 basesOn or off: DUST for DNA, SEG for proteinOff, the default
Genetic codeblastx, which reads the DNA query with it, and tblastn, which reads the stored DNA1, StandardThe 26 codes BLAST has, 1 to 6, 9 to 16, 21 to 31 and 33, named in the listNot used: the demo runs blastp

Use demo data, under the settings, fills the form with GLP-1(7-37), a peptide of 31 residues: residues 98 to 128 of UniProt P01275. It searches Patent proteins alone, so its result lists the settings. BLAST gives its short-query settings to protein of 30 residues or fewer, so by default this peptide runs with BLOSUM62 and word size 3. The demo sets PAM30 and word size 2, the short-query settings. Run it, then set both back to Default and run it again to see the hits and their scores change.

A value the program does not take stops the search before it starts, and the form says why, such as "Word size must be a whole number from 2 to 7 for a protein or translated search." Gap costs are checked for each query, since a short query runs with other scores: a pair that BLOSUM62 takes can fail for a protein of 30 residues or fewer, which BLAST scores with PAM30 unless you choose a matrix.

  • The FASTA36 methods take a matrix and gap costs, and have no word size, filter or genetic code. A gap costs its opening once and its extension for each letter, as in BLAST.
  • A matrix scores alike in BLAST and FASTA36: FASTA36 reads BLOSUM62 by its own name and each other matrix from the one BLAST uses under that name. BLAST's BLOSUM80 is on another scale from the BLOSUM80 file NCBI publishes, and Sequences uses BLAST's. The one difference is X: the FASTA36 methods score it as 0 under every matrix (see Search methods).
  • The search record lists every setting each query ran with, defaults included, and the ones you chose. Results for each query shows them in short, such as "PAM30 · gaps 9/1 · word 2 · filter off".
  • Run again runs with the same settings, and so does a follow-up search with the same method, such as the same search keeping more hits. A follow-up with another method starts at its defaults.
  • Query Match uses the scoring the search ran with: see Results. A query found whole reads 100 under any matrix, and one alignment reads differently under two matrices, so compare Query Match across searches with the same settings.
  • Align exactly scores with the FASTA36 defaults, BLOSUM62 at 11/1 for protein and 12/4 for DNA, whatever the search used.
  • A program sends the settings as matrix, gapOpen, gapExtend, wordSize, lowComplexity and geneticCode, which GET /api/openapi.json describes. See Programs and API tokens.

The automatic choice

Automatic is the first method on the list, and the search record says what it ran.

  • Protein to protein runs Smith-Waterman. Where FASTA36 is not installed, it runs BLAST, and the search says "FASTA36 is not available here, so this protein search used BLAST."
  • DNA to DNA runs BLAST. A query under 50 bases also gets an exact scan of every stored sequence on both strands, and the two are merged, so its exact matches lead the list and are complete.
  • DNA to protein runs BLAST, and a query under 50 bases gets a warning: "A DNA query under 50 bases is short for a protein search, so a match can be missed."
  • Protein to DNA runs BLAST.

Choosing a search method says why Automatic picks as it does.

Exact and pattern search

Exact text and Pattern read every stored sequence, so their counts are complete. They compare DNA with DNA or protein with protein. A DNA search covers both strands unless you clear Search both strands.

To try them, choose Use demo data. With Exact text it fills SEQ ID NO: 1 of US12788009B2, which its claim 1 recites (us12788009b2-seq1.fasta), or for DNA / RNA the probe of US12788762B2 (us12788762b2-probe.fasta). With Pattern, the button under Pattern fills claim 1 of US12788009B2 as a PROSITE pattern (us12788009b2-pattern.txt) or as a regular expression (us12788009b2-pattern-regex.txt), and for DNA / RNA a primer of US12788762B2 with IUPAC codes (us12788762b2-primer-pattern.txt).

Pattern syntax says how you write a pattern: PROSITE, the first choice, or Regular expression. Sequences never guesses, since [AG] and parentheses mean different things in the two.

A PROSITE pattern is parts joined by hyphens:

PartMeans
ACGTThese letters, in this order
[AG]A or G
{AG}Any letter but A or G
xAny letter
x(2,6)2 to 6 of any letter. A repeat goes on one letter or one bracket, as in A(3) or [AG](2,4)
< and >Before the first part, the start of the stored sequence; after the last, its end

For example, AAGCTT-x(2,6)-GGATCC finds a HindIII site, 2 to 6 letters, then a BamHI site.

A regular expression takes these parts:

PartMeans
ACGTThese letters, in this order
[AG] and [^AG]A or G, and any letter but A or G
.Any letter
{3}, {2,6} and {2,}3, 2 to 6, or 2 to 1,000 of the letter, bracket or dot before it
?, + and *0 or 1, 1 to 1,000, and 0 to 1,000 of the letter, bracket or dot before it
^ and $The start and the end of the stored sequence

Parentheses with a bar give choices: (GGATCC|AAGCTT) finds a BamHI or a HindIII site. An expression takes up to 8 choices, and two sets of choices in a row multiply, so (A|C)(G|T) makes 4. AAGCTT.{2,6}GGATCC finds what AAGCTT-x(2,6)-GGATCC finds. Sequences turns each choice into a PROSITE pattern and searches them together, so a place that two choices fit is listed once, and the search record keeps the expression as you typed it and its PROSITE form. A regular expression cannot hold a backreference such as \1, a lookahead or lookbehind such as (?=, or a repeat on a group such as (AG){3}, which you can write out as AGAGAG. The message names the part it cannot take.

In DNA, IUPAC codes such as R and N stand for their bases. A pattern takes at most 100 parts, 4 ranges such as x(2,6) or {2,6}, repeat counts up to 1,000, and 20,000 characters. In a regular expression ?, + and * are ranges too, a run of letters is one part, and each choice keeps within these limits.

Unless you turn on stored codes, a letter in a stored sequence matches only that letter, so an N in stored DNA, or an X in a stored protein, is not a wildcard here. An alignment counts it apart from a mismatch (see Results). Each row is one sequence on one strand. Occurrences counts every place the text or pattern fits, and First positions lists up to 20 of them, counted from the start of the stored sequence. The list stops at 1,000,000 rows, and the totals still count every match. If a search's rows and their alignments pass 8 GiB, or another size the server sets, the search fails with "The result exceeded its size limit. Use fewer or longer queries." The scan reads the set in parts, one for each thread, and each part stops keeping rows once its rows reach an equal share of that size, so the list can also end before 1,000,000 rows. The totals still count every match.

A DNA query with IUPAC codes reads differently by method. Pattern reads R as A or G, and each other code as the bases it stands for. Exact text matches R only to R, and BLAST, alone or in Automatic, counts R against A or G as a mismatch. So when the query box holds DNA with codes and the method is neither Pattern nor Within k changes, which reads codes as Pattern does, a note under the box offers Search as a pattern. It puts the query in Pattern as PROSITE letters, or up to 8 queries as one regular expression, and leaves the query box as it was. Unless you turn on stored codes, a pattern takes a code written in a stored sequence as a letter of its own, not as the bases it stands for, so the two searches find different things. In our check of the demo primers of US12788762B2 against two coronavirus reference genomes, Pattern found each site the codes allow and Exact text only the patent's own listing. BLAST found both, but read SEQ ID NO: 1 at 92.3% identity both where its codes fit and where one letter differs.

Stored IUPAC codes

A sequence listing can write a place that holds one of several bases or amino acids as an IUPAC code, such as R for A or G, N for any base, or X for any amino acid. Check Let stored IUPAC codes match the letters they stand for, under Exact text or Pattern, and a code in a stored sequence matches each letter it stands for: R matches A or G, Y matches C or T, N matches any base, and so on for each code. In a protein, B matches D or N, Z matches E or Q, J matches I or L, and X matches any amino acid. Most codes in one match, 3 to start, limits how many codes one match may hold to 1, 2, 3, 5 or 10, so that a run of N does not match every search. A code in an Exact text query still matches only the same code, and that match is not counted.

Whatever a search finds with the box clear, it finds with the box checked. A place counts when the search fits it with no more codes than Most codes in one match allows. Where a sequence has more than one such place, or a pattern fits one place in more than one way, the row shows the match with the fewest codes, and Occurrences and First positions count every place. The Stored codes column, shown for such a search, gives the number of codes in the row's match. The pane names each one as the alignment shows it, with its position in the stored sequence and, for Exact text, the letter of your query it stands for, such as "1 code: R at 7 for A". An Exact text row counts each code as a letter that is not identical, so a match of 8 letters with one code reads Identity 87.5% and leaves Query Match empty, while Identity, X as a match still counts a stored N or X as identical. In a pattern, x and . take a code as they take any letter, so it is not counted. The search record says the box was checked and gives the limit, and the CSV gives the number of codes in stored_codes.

Within k changes

Within k changes lists every stored sequence that differs from your query by at most k changes, which fits a claim such as "up to 3 substitutions". It reads every stored sequence of the set and applies no score, E-value or rank cut, so its counts are complete. It compares DNA with DNA or protein with protein.

Choose k, from 0 to 5, and what counts as a change:

  • Substitutions only: a letter changed in place. The match is a stretch of the stored sequence as long as your query, with at most k letters that differ.
  • Substitutions, insertions and deletions: your whole query, aligned anywhere in the stored sequence, with at most k changes of the three kinds in all.

An N, an X or another IUPAC code in your query matches any letter it stands for, so it is never a change. In a stored protein, an X or another code also matches any letter it stands for. Stored DNA is read letter for letter: a stored N or other code matches only the same code in your query, so against any other letter it counts as one change. A DNA query is searched on both strands, and RNA U is searched as T.

Each row is one stored sequence on one strand, with its closest match to your query. Changes counts that match's substitutions, insertions and deletions, and the list goes fewest changes first. The detail pane gives each kind of change, the Wildcard matches, the identity and the alignment. In the alignment, a wildcard match shows as a differing letter that is not a change. A stored sequence with a long run of X can match a query with no change at all, so rows with as many changes go fewest wildcard matches first, and a match letter for letter comes before one made of X. The CSV export adds the columns changes, substitutions, insertions, deletions and wildcard_matches. On Patents, families and publications sort by their best identity.

A query takes at most 64 letters, and k must be below its length. Each query lists up to 50,000 hits, a stored sequence on one strand each. A query with more lists its 50,000 closest: the fewest changes, then the fewest wildcard matches, then the set's order. It says how many stored sequences it found, and names the largest smaller k that would list every hit, or asks for a longer query when even k = 0 finds more. Its counts by number of changes stay complete in the search record.

To find its closest, a query with more than 1,000,000 hits reads the set a second time, so it takes up to twice as long, and it sorts at most 1,000,000 hits. When more than that have as few changes as its 50,000th closest or fewer, the list holds only the hits with fewer changes, or, at no change, the closest of the first 1,000,000 in the set's order, and the query says so.

Within k changes runs only where its scanner, kscan, is installed, and the Sequences container image includes it. Where kscan is missing, the method is greyed out with a note that says so.

Results

A finished search opens its results on a page of their own (see Addresses), 50 rows to a page, in tabs. Which tabs show depends on the search:

  • Patents: one row for each patent publication that holds a matching sequence, in a search of the snapshot.
  • Applicants: one row for each first-named applicant of those publications. It shows with Patents.
  • Alignments: one row for each alignment, in every search. An exact or pattern search calls it Matches.
  • CDR matches: the results of an antibody search, in place of Patents and Applicants.
  • Variants: the first hits of one query in the query's own positions, and what differs at each. See Variants.
  • Search record: what the search used and returned: the search engine version, the set and snapshot it searched, every setting, and the SHA-256 of each file the search wrote. A SHA-256 is a fingerprint of a file: it changes if the file changes at all.

A search opens on Patents when it found a patent publication. One that found none, such as a search of Swiss-Prot, opens on Alignments, with the Patents tab beside it. An antibody search opens on CDR matches.

The top of the page gives the search's name and, under it, what it searched. Its header holds Edit and run again, Run again, Copy link, Export matches (CSV) and Download search record, and Watch this search for a search that can be watched. Edit and run again opens the search page with the form filled in as the search ran: its queries, sets, method, search settings and name, or an antibody search's CDRs. Change what you need, then run it. Run again starts the same search at once; see Run again. A search with no Run again, such as one a case's plan started, has no Edit and run again either. A printed result page holds the result and its sources and terms, without the page's menus and buttons.

Each row of Patents and Alignments is one line. Select a row to see all of it in its pane: on a wide screen the pane sits beside the list, and each scrolls on its own; on a narrower screen the pane opens under the row. The pane holds the row's measures, its marks and its alignment. These keys work while the list has the focus, and Keys, above the list, lists them too:

  • Up and Down move through the list. At the end of a page they open the next or previous page.
  • Page Up and Page Down move 10 rows. Home and End go to the first or last row.
  • K keeps and X excludes the selected row, and the next row is selected. U clears its mark.
  • R goes to the reason field, and Esc comes back to the list.
  • C copies the alignment.

A list that does not load says so in place of its "Loading" line, such as "The patents did not load.", with the server's reason when it gave one. Try again reads it again.

The line under the search's name says what it searched and the date of that set's source. Results for each query, above the tabs, gives each query's length, program, E-value cutoff and number of sequences returned. A query that reached its limit says so. A BLAST query's line reads "This query reached the 5,000-sequence limit. More matches exist." A FASTA36 search (ssearch36, ggsearch36 or glsearch36) keeps the best-scoring sequences at its limit, and its line reads "This query reached the 5,000-sequence limit. The list holds the best-scoring 5,000. For every exact occurrence, run the query as Exact text."

Each measure has one name in the list, its filters and sorts, the pane and the PDF report, one column in the CSV and the workbook, and one field in the API. Each sorts Alignments and sets a minimum under Filters; Count an X as a match holds Minimum identity to Identity, X as a match.

MeasureWhat it countsCSV and workbook columnAPI field
Query MatchThe alignment's score as a percent of the query's score against itself, both under the matrix, or DNA scores, and the gap costs the search usedquery_matchqueryMatchPct
IdentityIdentical residues as a percent of the alignment's length, gaps includedidentityidentityPct
Query identityIdentical residues as a percent of the full length of the queryquery_identityqueryIdentityPct
Hit identityIdentical residues as a percent of the full length of the hithit_identityhitIdentityPct
Query coverageThe span of the alignment as a percent of the full length of the querycoveragecoveragePct
Identity, X as a matchIdentity with each X in a stored protein, or N in stored DNA, counted as identical. The pane shows it where an alignment has oneidentity_x_as_matchidentityXPct

Query identity and Hit identity answer different questions. A short stored sequence found whole inside a long query reads Hit identity 100 and a low Query identity; a short query found inside a long stored sequence reads the reverse.

Each method gives the measures its matches can count:

MethodQuery MatchIdentity, Query identity, Hit identityQuery coverage
Alignment (BLAST), blastn and blastpYesYesYes
Alignment (BLAST), blastx and tblastnNoYes. A blastx query is DNA, so each identical residue counts three of its letters; a tblastn hit is DNA, so three of the hit'sYes
Smith-Waterman, Global on the query, Global on both sequencesYesYesYes
Exact text100Identity and Query identity are 100; Hit identity is the query's length as a percent of the hit's100
Within k changesNo: it has no scoreYes100
PatternNoNoNo

An Exact text search with stored codes counts each code in a match as a letter that is not identical, so a row whose match holds a code reads Identity below 100% and leaves Query Match empty.

A sequence listing writes X for any amino acid and N for any nucleotide, often where its notes describe the residue, such as a modified one. Where the stored sequence has one facing a letter of your query, the alignment marks it apart from the mismatches, with a legend under the counts, and Identity gives the count in words, such as "96.6% (28 of 29 identical, and 1 X)". Identity, X as a match counts each one as identical, such as "100.0% (29 of 29)". In a tblastn search the X is in the stored DNA's translation, from a codon with an N or another letter for more than one nucleotide. The snapshot does not keep the listing's notes, so the pane says so and links to the publication's page: the listing filed with the publication may say what each one stands for. Where the server runs Modifications, the hit's pane also lists the residues its records mark as modified, which may name what an X stands for. Sequences counts these letters from each alignment when it opens the result. The score and the other measures stay as the search program gave them, and the Search record says how the letters were counted.

How to read a result works out Query Match, Identity, Query identity and Query coverage for one hit of a real search.

Charts

Charts, above the tabs, draws the hits three ways. Each chart reads the rows the list reads, with the same filters, so a change to a filter changes the charts too. Under each chart, Show table gives its numbers as a table, and Download SVG and Download PNG save it as a picture.

  • Identity against coverage: a dot for each hit, placed by its identity and its query coverage. Hits that fall in one place share a dot, which grows with them. With a Published before date, each dot takes the color of its date group: before your date, filed before it, later, or with no date. The chart keeps the later hits that the list leaves out, so you can see what the date takes away. Minimum identity and Minimum query coverage show as dashed lines, and the hits that fall short of them turn grey.
  • Hits by year: the publications, or the families when Group says Families, by the year each was first published, stacked by date group, with your date marked. Those with no date stand apart at the right.
  • Where hits align on the query: the query along the top, and a bar for each of the first 20 hits in the order Sort matches gives, over the part of the query its alignment covers. Ticks mark where the hit differs from the query. A ring marks an X in the hit's sequence (N in DNA), the letter a listing uses for any residue, so it reads apart from a real difference.

Drag across Identity against coverage to keep only the hits in that area. The list, its counts, the other charts and the CSV export follow the area until you choose Clear area. With the keyboard, the arrow keys move a cursor over the chart (with Shift, 5 points at a time), Enter sets one corner and then the other, and Delete clears the area. On the other two charts, the arrow keys step through the years or the bars, and Home and End go to the first and the last.

A combined search and an antibody search have no charts. To try them before you sign in, open a sample result at /samples: each sample draws the same charts, and a drag there opens the sample again with your area.

By patent

Group, above the list, sets what a row of the Patents tab is. With Families, the first choice, each row is one patent family: the publications that share a priority claim or an application, or, where no priority claim is on record, the same title and first applicant. On a server that runs Patent families, the families that Google Patents Public Data gives come first. The row shows the family's earliest publication, with its number, title and applicant and when it was published, the family's best Query Match and Identity, and how many sequences match, each counted once. A chip such as "+2 in family" counts its other publications, and the line over the list counts both, such as "12 publications in 5 families match these filters." A family of one publication reads as that publication. With Every publication, each row is one publication.

Three more choices of Group keep one row for each publication and put the rows under a heading for each group: Applicant, with names joined as on the Applicants tab and the best identity first; Patent office, the office with the most publications first; and Year published, oldest first, or newest first when Sort patents is Latest published. Each heading gives its count, such as "US, 3 publications", and the line over the list counts the groups, such as "10 publications in 5 patent offices, 1 with no office, match these filters." Publications with no applicant, office or date come last. Within a group the rows keep the sort, and Up and Down pass over the headings. To try it, open a sample result at /samples.

A publication's pane opens with a line that says why it is listed, built by fixed rules from the result. It says how many of the publication's sequences match and how the best one compares with your query, such as "is identical to your query over all 31 letters" or "has 28 of 29 aligned letters identical, and the other is an X". With a date in Published before, it says where the publication falls against it. And it says how many other publications in the result hold all the letters of your query unchanged, with the earliest of them, or that none is earlier. Holding your query unchanged means an alignment over the whole query with every letter identical; it says nothing about the rest of the stored sequence, which can be longer. The line never says that a publication anticipates a claim or is prior art: you decide. The Patents CSV has the same line as why_listed.

A publication's pane links its number to its page at Google Patents, with the number written as Google Patents writes that office's numbers in that era: a Japanese number from before 2000 by its era year, for example, and a Korean one with a two-digit year before 2000. Where Google Patents does not have the publication, as with Korean publications of international applications, the number links to Espacenet and the pane says "Google Patents does not have this publication." A US publication from the last six weeks may not be on Google Patents yet, and the pane says so. Also at, under the title, links the publication at Espacenet and, for a US publication, at Patent Public Search, the US patent office's search. A server can set another address for the number's link. Where a publication's record gives its filing date or priority claims, as a US published application's does, the pane shows them beside the publication date, such as "Filed 2026-06-09" and "Priority 2024-01-31". Many records in the patent sets give neither. The pane lists the publication's matching sequences by SEQ ID NO, with the best match's alignment under them; choose another to see its alignment. On a server with patent context, What the patent says follows.

A family's pane says how the family was formed: "Grouped by a shared priority claim," with the claim's number, "Grouped as publications of one application," with the application's number, "Grouped by the same title and applicant. No priority claim is on record," or, on a server that runs Patent families, "Grouped by Google Patents Public Data as family" with the family's number. It lists the family's publications, earliest first, each with its dates and an Open link to its page. It lists up to 20 and counts the rest; Every publication shows each one. Under them are the earliest publication's matching sequences, with the best match's alignment; choose another publication to see its own.

Rows that tie on the sort go earliest first, undated last, so among the rows with the best Query Match the earliest is on top. Published before tests each sequence by its earliest date and by the filing and priority dates of its US and WO publications, and folds the later publications under the list (see Dates). Applicants and a case's report count publications, whatever Group says.

What the patent says

Whether you keep a patent can turn on what it does with the sequence: claims it, makes it, tests it, or only lists it. On a server that holds patent context, a US publication's pane has a part called What the patent says, under its matching sequences, that quotes the publication on its matching SEQ ID NOs:

  • A line for each matching SEQ ID NO, from fixed rules, such as "SEQ ID NO 1 is named in claims 1 and 3 and in paragraphs [0012] and [0040]." A number named inside a range, such as SEQ ID NOs 1 to 40, counts as named.
  • Abstract: the publication's abstract.
  • Claims: each claim that names a matching SEQ ID NO. Where none does, the first claim, labeled so.
  • Description: for each matching SEQ ID NO, the first three paragraphs of the description that name it, with their numbers. The publications of 2001 to 2004 in the older grant form give their abstract and claims only.
  • Sequence listing notes: the notes of the publication's sequence listing for those sequences: each feature's key, where it lies and its note, such as what an X stands for.

Each quote names its claim or paragraph and links to the publication's full text. A quote is the document's own words, never a summary. A very long claim or paragraph is cut at a word, with "Cut here; the full text goes on." under it. The quotes come from the US patent office's weekly full-text files that the snapshot was built from, kept in a folder on the server, so a search takes no longer and no other site is asked. The part quotes and counts; it does not judge what a document means for a claim.

A publication of another office shows "Quotes come from the full text of US publications. Open this one at Espacenet for its text." Where the server's staff have turned on other offices, it shows the publication's English title and abstract from EPO's Open Patent Services instead, with Espacenet for its claims and description. A server without patent context leaves the part out. Its staff can add it: see Patent context in the installation guide. In a family's pane, the part is for the member you choose. In a case, the References step shows the same part in a reference's open row.

Compare, at the end of a publication's pane on Patents, adds the publication to a comparison of up to 3. Once 2 are chosen, Compare side by side opens them in columns: each publication's best alignment to your query, then, on a server with patent context, what it says. Choose Compare again to take one out.

By applicant

On the Applicants tab, each row is one applicant, the first applicant each publication names, with its publications, its earliest date and its best Identity. Rows are sorted by best Identity, highest first, then by earliest date, earliest first. An exact or pattern search has no identity, so its rows are sorted by earliest date. Publications that name no applicant come last, under "Applicant not given". The filters work as on Patents, and with Published before set the later publications fold under the list as they do there (see Dates).

  • Spellings that differ only in case, accents, punctuation or a legal ending such as Inc or GmbH share a row, and the row lists the other spellings.
  • To treat names that still differ as one, such as a company's old and new names, select their rows and choose Join selected names. A join takes 2 to 50 names and covers all of your searches until you choose Undo join on its row.
  • Join log, under the table, lists the joins and undos on your account, newest first: the newest 50, with the total when there are more. The log keeps every one. Each event holds the SHA-256 of the one before it, and Check the log reads every event to show whether any was changed or removed.
  • Where an applicant's name matches an organization's name in the Research Organization Registry (ROR), the row links to that record under Suggested ROR record. A suggestion never joins or renames names. A server shows suggestions only when its staff have set them up.
  • In a search of the snapshot, each row with an applicant's name has Watch this applicant and Export portfolio. See Watches and Portfolio export.

By alignment

Each row is one alignment of a query with a stored sequence: the hit, its title, and how many other publications hold the same sequence, such as "+4". A search of more than one query adds a Query column. A patent hit's title names the sequence's earliest publication. First published gives the hit's earliest date, or "Undated". Species gives the organism the hit's records name most often, such as Homo sapiens, and a count such as "+2" when they name others. See Species.

The pane gives the hit's measures and the alignment itself, 60 letters to a line, with each letter of the hit that differs from the query in bold. Copy alignment and Download .txt give the alignment as text. A patent hit's pane gives the earliest publication's applicant, with its filing date and earliest priority claim where its record gives them, and lists where the sequence is published: the earliest five, then Show all for the rest, 50 at a time. Then come up to five database entries that hold the sequence, linked to their records, and up to three papers, earliest first. See Papers. The pane lists each organism the records name, with its NCBI taxon, linked to NCBI's Taxonomy Browser, its group and how many records name it. On a server that runs clinical context, a protein hit's pane can also have a Clinical context part. See Clinical context. On a server that runs the funding trail, a paper with grants in PubMed has a Funding part under it, and a hit can show its US published applications' statements on federally sponsored research after its papers. See Funding trail. On a server that runs known variants, a hit identical to a human Swiss-Prot entry also shows the query's changes and the protein's ClinVar variants. See Known variants.

Columns

Columns, above the list on Patents and on Alignments or Matches, sets which columns the list shows and in what order. Tick a column to show it, clear it to hide it, and move it with its arrows. The name of the hit or patent stays first and Mark stays last. Alignments can add hit identity, E-value, bit score, score, alignment length, mismatches, gaps, query and hit positions, hit length and accession, and, in a patent set, the applicant, the publications and offices, and the filing and priority dates. Matches of an exact text search can add hit identity too. Patents can add office, kind, applicant, filing and priority dates, application, alignments, SEQ ID NOs, family size and date group. Your choice is kept for your account, or on a server without sign-in for that server, and applies to every search; Reset columns goes back to the standard columns. On a narrow screen the list hides some columns, as it does the standard ones. A sample result has the same choice under Columns, kept in its address, so you can try it at /samples.

Transform

Transform, above the list on Alignments or Matches, changes the list of a finished search. Each change is kept with the search and shows at once, and Undo takes it back.

  • Fold the same alignment makes one row of the hits that align the same way, the same query letters against the same hit letters: the one first published, with a chip such as "+2 same alignment" for the others.
  • Count an X as a match reads an X in a protein hit, or an N in a DNA hit, as the query's letter. Identity and query identity are counted again from the alignment, and Query Match too on a protein search. Each row it changed has a chip such as "2 X counted". Exact text and pattern matches have nothing to count.
  • Keep hits that cover query positions keeps the hits whose alignment covers every position from From to To.
  • Keep hits that differ from the query at positions keeps the hits with another letter or a gap at any of the positions you list, such as 3, 7, 12, or, with Every one, at all of them. A position outside a hit's alignment does not count.
  • Keep hits another search also found keeps the hits whose sequence another finished search found too, matched as Compare searches matches them. For every sequence that any of several searches found, choose In any search there.

Each change in force says what it does to the whole list, such as "6 of 7 hits cover query positions 1 to 20.", and the line over the list counts the changes. Export matches (CSV) and the Alignments sheet of Export all tabs (Excel) save the list as it shows. Patents, Applicants, a patent's own hits and a case report keep every hit the search found. Search record lists each change and each undo with its time, and so does the workbook's Search record sheet; the search's own record and its SHA-256 never change. Fold the same alignment, Count an X as a match and Keep hits that differ from the query at positions read every alignment, so a very large search does not offer them.

To try it, open Alignments on a search of your own. Or choose Use demo data under Compare searches, then Open results on the BLAST search and Alignments: fold the list, or keep the hits the Exact text search also found. A sample result is read-only: its list stays as its search left it.

Variants

Variants, beside Alignments, draws the first 100 hits of one query in the query's own positions, from the alignments the search kept. It takes the list's filters, sort and changes, and leaves out the hits you excluded. A search of more than one query has a Query choice.

Each row is one hit. A dot is the query's letter, a letter is the hit's own where it differs, and a dash is a gap in the hit. An underlined place has extra letters after it. Under the query, Changes marks each position where hits differ with the change most of them make, in a stronger fill the more of them differ. The grid shows as many positions as fit the screen, and Previous and Next move along the query.

Pick a position on the grid, in Position or under Most changed positions, to see how many of the hits that cover it differ, and each change with its count: K45R for another letter, K45del for a gap and 45insGS for extra letters after it. Each hit that differs there opens in Alignments, selected as a click on its row does. On a server that runs known variants, the pane of a hit identical to a human Swiss-Prot entry also shows the protein's ClinVar variants. See Known variants.

Export positions (CSV) saves a line for each query position: its letter, the hits drawn, the hits that cover it and the hits that differ, and each change and run of extra letters with how many hits have it. Programs read the same with GET /api/jobs/{id}/variants and GET /api/jobs/{id}/variants.csv.

A search that translates the query, such as blastx, and a pattern search have no letter-for-letter positions, so Variants says so and draws nothing. A hit whose alignment is not on record, as in some result files, is counted and not drawn. Letters a hit holds before the first query letter it aligns or after the last are not counted.

Papers

Where the server has PubMed's records, as the hosted service does:

  • Each paper reads as a citation: up to three authors, the title, the journal, the year with volume, issue and pages, and the day it went online if that came before its issue. Its PubMed ID and DOI follow, each linked.
  • A paper that was retracted, drew an expression of concern or was reinstated says so in a line under its citation, with the day, the notice and the source: PubMed, the Retraction Watch Database, or both. When no source gives the day, the line says "notice date not given". Corrections are not marked, and a mark never moves or hides a paper.
  • A PubMed ID that the National Library of Medicine deleted keeps its year, and a line says the paper is not in PubMed.
  • A paper that PubMed does not hold, but that has a DOI, adds the day or month Crossref gives for it, as in "Published 2021-12-01 (Crossref).", where the server has Crossref's dates. When Crossref gives only a year, the paper keeps its year.

Without PubMed's records, a paper shows its title and year, linked to PubMed, or to its DOI when it has no PubMed ID, and carries no marks. A search keeps its papers as they were when it ran. To see newer citations and marks, use Run again.

Species

A record of a sequence names the organism it came from, as its source gives it. Sequences places each record in one of these groups, with its NCBI taxon where it has one:

GroupHolds
Human, Mouse, RatHomo sapiens, Mus musculus and Rattus norvegicus, each with the taxa NCBI places under it
Other rodents, Other primatesThe rest of the rodents, and the primates other than humans
CamelidsCamels, llamas, alpacas and the rest of NCBI's Camelidae
Other mammals, Other animalsThe rest of the mammals, then the rest of the animals
Plants, Fungi, Bacteria, VirusesEach as NCBI's taxonomy holds it. Plants are its green plants (Viridiplantae)
OtherA taxon in none of the groups above, such as an archaeon or a protist
Artificial or syntheticA made sequence: NCBI's artificial sequences, such as synthetic construct, and records whose organism is a name such as "Artificial Sequence"
Not statedA record that names no organism
Not matchedA record whose organism matches no NCBI taxon, or matches two

A record takes the NCBI taxon its source states. A record that states none takes the taxon whose NCBI scientific name or synonym is its organism, apart from case and spacing. A name with its common name in brackets, such as "Homo sapiens (human)", takes the taxon both parts name. Sequences never guesses a name it cannot match: the record reads Not matched, and the pane gives the name as the record states it. A search keeps the species it had when it ran. A search from before species has no groups: run it again to filter it by species.

Clinical context

Clinical context is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Clinical context section describes. Where it runs, a Swiss-Prot or TrEMBL hit whose UniProt entry is a drug target has a closed Clinical context part in its pane on Alignments, with the target's name and how many drugs act on it. Opened, it shows:

  • The drugs that act on the target, highest clinical stage first, each with its mechanism and action.
  • The diseases each drug is studied for, each with the highest stage it reached.
  • Each disease's trials, ten at a time, with their phase and whether results were posted.

The target, drugs and diseases link to Open Targets, each drug to ChEMBL too, and each trial to ClinicalTrials.gov. The data come from Open Targets (with ChEMBL), ClinicalTrials.gov and AACT's copy of trial results, and Sources and terms credits them. The part names its sources and their dates, and it reports what they say without drawing a conclusion. Patent, EST and attached database hits show none, and neither does an antibody CDR search.

Funding trail

The funding trail lists the grants PubMed gives for each paper under a hit, with the NIH awards they match. It also shows the statements on federally sponsored research in the US published applications that hold a hit's sequence. It is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Funding trail section describes, and such a server has a funding trail card under Sources and terms.

PubMed lists each grant by its number. A grant shows as an NIH award when its number matches exactly one NIH project in NIH RePORTER, NIH's public database of research projects. An award here is one project under one project number. NIH funds a project year by year, so an award can span many fiscal years, each running from October to September.

On Alignments, a paper in a hit's pane has a closed Funding part when PubMed lists grants for it. Its label gives the counts, such as "Funding: 6 NIH awards, 1 other grant". Opened, it shows:

  • The NIH awards, in the order PubMed lists their grants. Each has its project number, linked to NIH RePORTER, the title and each institution with its fiscal years. Then come the investigators NIH lists for the latest fiscal year that has any, with their names as NIH writes them. The contact investigator comes first, and in many older records the name ends in "(contact)".
  • The other grants, as PubMed lists them, each with its number, funder and country.

Each list shows five, and the rest open under "N more". An award with many investigators shows six names and says how many more there are.

PubMed often lists one project in two ways, such as P30 CA016672 and P30-CA16672, and the award then shows once. All other grants show as PubMed gives them: grants from other funders, contracts, grants with no number, and numbers that match no single NIH project, such as an NIH grant older than the fiscal years NIH RePORTER covers. A number written with other words, such as ARRA R01HD065288, shows that way too, even when its project is among the awards. The part's first line gives the date of each source and the fiscal years NIH RePORTER covers.

The part shows what PubMed and NIH RePORTER list, and no award amounts. It does not show that a grant paid for the sequence, the work in the paper or a patent. The investigators are the people NIH lists for the award, not the paper's authors. A paper with no Funding part is one for which this server's copy of PubMed lists no grant, which does not mean the work had no funding.

A US published application whose invention was made with federal funding says so in its description. Where the server's funding trail holds these statements, a hit's papers are followed by a line, "Federally sponsored research, as the applications state it" ("as the application states it" for one), and a closed part for each application, named by its number and publication date. A hit shows up to three, from the earliest applications that hold its sequence. When more of them give a statement, a last line says how many, such as "2 more applications". Opened, a part shows the statement word for word, after the heading the application puts above it, if there is one. A statement that only says it does not apply, such as "Not applicable", is left out.

The statement is the application's own words, not matched to an NIH award or to a paper's grants. A hit without such a part may still come from federally funded work: its applications may not say so, or this server's copy may not hold the week they were published.

Known variants

Known variants is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Known variants section describes. Where it runs, a hit identical to a human Swiss-Prot entry shows the variants ClinVar records on that protein. For a protein query, it also shows where the query differs from the protein and whether ClinVar records each of those changes. ClinVar is NCBI's public archive of human genetic variants and what submitters report about each one.

ClinVar describes each protein change against a RefSeq protein, such as p.Val600Glu on NP_004324.2, BRAF's protein. Sequences keeps a ClinVar change only when that RefSeq protein is identical to the canonical sequence of a human Swiss-Prot entry and has the change's reference residue at the stated position. Every change is then in the entry's own numbering. A change ClinVar describes only against another isoform is left out, and no change is mapped from another sequence by alignment. A hit qualifies when its sequence is identical to an entry with at least one kept change, so an identical patent sequence qualifies too.

When a protein query hits such a protein, Alignments adds two columns. Known changes counts the query's changes that ClinVar records. ClinVar variants counts the protein's ClinVar variants. The hit's pane shows:

  • The query's changes, from aligning the query with the protein the same way Align exactly does. Each place where they differ is named in HGVS protein form, such as p.Val600Glu, in the protein's numbering. The list opens with how many of the query's changes ClinVar records, such as "Known changes: 1 of 2", and ends with the stretch of the protein the alignment covers, such as "From an exact alignment by ssearch36 over residues 1 to 766 of this protein."
  • Under each change found in ClinVar, one line per ClinVar record, such as "Known: ClinVar 13961, Conflicting classifications of pathogenicity, public since 31 Jan 2015" for BRAF's p.Val600Glu, linked to that record. Every other change reads "Not in ClinVar".
  • The first ten changes, with the rest under "N more changes". A change that holds X, or another letter that does not stand for a single amino acid, is counted in the opening line but not named, and a line under the list says how many there are.
  • A closed ClinVar variants part. Opened, it lists the protein's ClinVar variants 25 at a time, each with its change, its record, its classification and the day it became public. Clinical significance filters the list to one classification.

Classifications are ClinVar's germline classifications, and "No germline classification" marks a record that has none. "Public since" is the day ClinVar first made the record public, as the record's DateCreated in ClinVar's weekly XML release gives it. A record without that day shows no date. The ClinVar variants part names its sources and their dates, such as "From ClinVar of 8 Nov 2026, on UniProtKB/Swiss-Prot 2026_04 and RefSeq of 10 Nov 2026", and Sources and terms credits ClinVar, RefSeq and UniProt. On Alignments, Export matches (CSV) and the Alignments sheet of Export all tabs (Excel) give the same in three columns, known_changes, query_changes and clinvar_variants. A query_changes cell reads like "p.Val600Glu known ClinVar 13961; p.Ala598Thr not in ClinVar".

"Not in ClinVar" means that this release of ClinVar records no such change on this protein. It does not show that a change is new. ClinVar holds what submitters send it, a change may be on record only for another isoform, and a change can be published without ever reaching ClinVar. Known variants reports what ClinVar records and draws no conclusion.

A search keeps the known changes it found when it ran. When the server has a newer release of known variants, the ClinVar variants part names the release the search's changes came from. To check the changes against the newer release, use Run again. A DNA query, a pattern search and a Within k changes search list no query changes, and hits from an attached database show none of this. Changes in DNA, named on a human transcript, are not covered yet.

Genes

Genes is an add-on: it shows only on servers whose staff have turned it on, as the hosted service has. Where it runs, a new search shows the human gene behind two kinds of hit:

  • A Swiss-Prot hit on a human protein shows each gene its Swiss-Prot entry names. Of the 20,430 human entries, 20,256 have a gene this way.
  • A patent protein hit whose sequence is identical, letter for letter, to a human Swiss-Prot entry's shows that entry's genes.

A hit never takes a gene from a sequence that is only similar. TrEMBL, EST, patent DNA and attached database hits, and Swiss-Prot hits on other species, show none.

  • On Alignments, or Matches after an Exact text or Pattern search, the Gene column gives each gene's symbol. It links to the gene's page in ConductGraph's library, in a new tab, and that page needs a ConductGraph sign-in.
  • The hit's pane names the gene, as in "Gene: PCSK9", or the genes when an entry names more than one, such as a gene and its copy. Each has its full name, its NCBI Gene and HGNC IDs and how the hit links to it: "Swiss-Prot's entry names it", or "same sequence as" and the Swiss-Prot entry. A last line names the sources and their dates.
  • A search keeps its genes as they were when it ran, so one run before genes was turned on shows none. To see them, use Run again.

An antibody search's CDR matches, a combined search's References and Compare searches show no genes; the hit list and pane of each search do. The genes come from NCBI Gene and HGNC, linked through UniProtKB/Swiss-Prot in ConductGraph's Gene search, and Sources and terms has a card that credits them.

Keep and exclude

Each row of Patents and Alignments can be marked Keep or Exclude in its pane, and Clear takes the mark off. Before you mark a publication, the line at the top of its pane says why it is listed (see By patent). The keys K, X and U do the same from the list, and after K or X the next row is selected. While a row's mark saves, another key or button does nothing to that row, so K pressed twice keeps it once. A search of a collection you imported has no marks. Type a reason in Reason (optional) before you mark the row, and it stays with the mark. To give a marked row a new reason, type it and press Enter. The list's Mark column says "Kept" or "Excluded". Excluded rows leave the list until you choose Show excluded. The CSV files keep every row that passes the filters, excluded ones too, with the marks and reasons. A result opened from a case records each mark in the case's log, and the case's report lists the publications and matches you keep.

On Patents with Families, a mark on a family row goes on each of its publications in the list at once, but not on later ones the date folds, and a case's log records it as one event that names them all. The Mark column says "Kept" when every publication of the family is kept and "Some kept" when only some are; the pane says how many, such as "2 of 3 kept." A family's mark is saved 500 publications at a time. If a part does not save, the list shows the marks that did, the family stays selected, and the page says how many, such as "Only 500 of the family's 1,000 publications were kept." Mark the family again to finish. To mark one publication of a family, choose Every publication. A publication you exclude leaves its family, and Show excluded lists it.

Align exactly

After a BLAST search of a set in the snapshot that compares like with like (blastn or blastp), and where FASTA36 is installed, each row has a Select box. Select up to 1,000 rows, or choose Select this page, pick Smith-Waterman (local) or Global on the query under Exact alignment, and choose Align exactly. The whole query of each selected row is aligned again with the stored sequence, and the new scores join the row and the CSV. Global on the query aligns every letter of the query, so its scores cover the query's full length.

Compare searches

Compare searches, below Recent jobs on the search page, shows which sequences two to six finished searches found in common and which only some of them found. Tick the searches and choose Compare: they are numbered in the order you ticked them. Each sequence falls in one part, such as Only in search 1, In both or In all 3, and each part shows its count. Choose a part to list its sequences, 50 a page, each with the best measure each search gave it, such as its Query Match. In any search, after the parts, lists every sequence any of the searches found, each with every search's measure or Not found. Open results opens any of the searches.

Searches of the same set match a sequence when they found the same stored sequence. Searches of different sets or snapshots match it by its letters, and searches of an attached database match only within that database. A result file imported into a case counts as a search of the set its rows were looked up in, when that is one set. A search of a database you imported under Import a database, or one made before 0.4.0, keeps its results only in its full record, so it cannot be compared.

Choose Patents, under the searches compared, to see the same searches by patent. Each patent any of them found is listed once, with the searches that found it and, for each search, the SEQ ID NOs it hit and its best Query Match and identity, Not found, or Excluded where you excluded the patent on that search's Patents tab, which then does not count it. The parts work as the sequence parts do, counted in patents, and the list opens on In any search, every patent the searches found. Searches share a patent when its office, number, kind code and publication date are the same, even across sets and snapshots, so an A1 and its B1 are listed apart. Minimum identity (%), Published before and Patent office narrow every search's patents alike. Download CSV and Download Excel save every patent of the part shown, as filtered. A search whose results name no patents, such as a search of an attached database, adds none. The list holds up to 200,000 patents; when the searches find more, set a higher Minimum identity or an earlier Published before date, or choose one Patent office.

To try it, choose Use demo data. It searches SIINFEKL, eight letters of chicken ovalbumin, on Patent proteins by Exact text and by BLAST, then opens the two compared: Exact text finds every stored sequence that holds the eight letters, and BLAST finds near matches too, so the parts show what each method finds and misses. Choose it again to open the same two searches, with no new search.

Drug names

Drug names is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Drug names section describes. Where it runs, a hit whose sequence is the same as one of a drug's chains names the drug in the Drugs column on Alignments, and at the top of the hit's pane, in a line such as "Drug: TRASTUZUMAB (CHEMBL1201585), same whole chain, highest stage approved". The same means one of three things:

  • The whole chain: the hit's sequence is the drug's chain, letter for letter.
  • A variable domain: the hit holds the drug's heavy or light chain variable domain whole, as ANARCI numbers it. A hit that holds both, such as a single-chain fragment, says so.
  • The bases: the hit's sequence is an oligonucleotide drug's strand, base for base, with U read as T. Sequences compares the bases alone. An oligonucleotide drug carries chemical changes to its sugars, bases and backbone, which are not compared.

A near match names no drug, however close. Each line gives the drug's highest stage as ChEMBL states it, or "not stated", and links to the drug's ChEMBL compound page or its FDA UNII record. Under the list, the pane credits the sources: ChEMBL, whose data are shared under the Creative Commons Attribution-ShareAlike 3.0 Unported license (CC BY-SA 3.0), and GSRS, the substance registry of the FDA and NCATS. Sources and terms has a card for each, with how we read its terms.

Drugs only, beside Patent office, keeps the hits that name a drug, and the CSV export follows it. A search names drugs as it runs and keeps them. A search from before the add-on ran names none; use Run again. Hits of an attached database name none. To search the drug chains themselves, check Drug proteins or Drug nucleic acids under Search in, where the server holds them (Sets).

Claimed sequences

Claimed sequences is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Claimed sequences section describes. Where it runs, it says which claims of a hit's US applications and patents name the hit's SEQ ID NO. The Claimed column on Alignments reads Yes when a claim of one of the hit's US publications names its SEQ ID NO, and No when the claims of each of its US publications were read and none names it. It is empty for a hit with no US publication, or one whose claims were not read. Point at Yes to see which publications and claims.

In the hit's pane, under each US publication whose claims were read, a line such as "Claimed: SEQ ID NO 1 (claims 1, 3; at least one independent). Not claimed: SEQ ID NO 10." gives the hit's SEQ ID NOs that its claims name, with those claims, and the ones no claim names. A claim names a SEQ ID NO when its text gives the number, on its own, in a list or in a range. A cancelled claim names none, and a claim that gives a sequence only by its letters is not counted. Fixed rules read each publication's claims as it prints them, from the US patent office's weekly full-text files.

Claimed only, under Claims beside Patent office, keeps the hits that a claim names. On Patents it keeps the publications whose own claims name a hit's SEQ ID NO. The CSV export and the workbook follow it. A search reads the claims as it runs and keeps what it found. A search from before the add-on ran shows none; use Run again. Hits of an attached database show none.

Patent families

Patent families is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Patent families section describes. Where it runs, Patents with Families groups publications by the families Google Patents Public Data gives first. Each of those families holds the publications that share the same priority claims. Publications it places in one family stay one family, and two of its families never join. A publication it does not place joins one of its families by the rules under By patent when its priority claim, application number, or title and first applicant link it to exactly one of them; otherwise the rules group it as they would without the add-on.

A family's pane then says "Grouped by Google Patents Public Data as family" with the family's number. Whole family gives how many publications Google Patents Public Data lists in the family and in which offices, and Other publications in this family lists the ones your result does not hold, up to 20, each linked to its page. Both start with publications from 2001, and the add-on places no earlier publication in a family. A search reads the families as it runs and keeps them, so its families stay the same when the server's copy of the data is replaced. A search from before the add-on ran groups by the rules alone; use Run again.

Patent status

Patent status is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Patent status section describes. Where it runs, it gives the status of each US application of a hit in the US patent office's words, with the office's status date. The statuses are copied from the office's Open Data Portal on one day, and may have changed since. Each tooltip, line and export gives that day.

The US status column on Alignments gives the group of each US application of the hit: "Granted", or "Granted 2, Abandoned 1" for a hit with three. The numbers count applications, so a patent and the published application it came from count once. Point at the cell to see each US publication with the office's words and status date. On Patents, the US status column gives the group of the publication's own application. In the hit's pane, a line such as "US status as of 2026-11-17: Patented Case, status date 2018-05-30." follows each US publication.

GroupThe office's words
GrantedPatented Case
PendingWords of an application in pre-examination, examination, allowance or appeal, such as Non Final Action Mailed
Expired (fee unpaid)Patent Expired Due to NonPayment of Maintenance Fees Under 37 CFR 1.362
AbandonedWords that begin Abandoned, Abandonment or Expressly Abandoned
OtherAny other status, such as Court Proceedings Terminated

Granted means the office's status read Patented Case on the day shown. The office keeps that status after a patent's term ends, so a granted patent may no longer be in force. Sequences does not estimate when a term ends. No status shows for an application the portal does not hold, such as one filed before 2001, for a publication with no application number, or for another office's publication.

US status, beside Claims, keeps the hits with at least one US application in the group you choose, so a hit it keeps can show other groups too. On Patents it keeps the US publications whose own application is in it. The CSV export and the workbook follow it. A search reads the statuses as it runs and keeps them. Run again uses the copy the server holds now. A search from before the add-on ran shows none. Hits of an attached database show none.

Modifications

Modifications is an add-on: it shows only on servers whose staff have turned it on, as the installation guide's Modifications section describes. Where it runs, it shows the residues and bases that a hit's records mark as modified, in the records' own words, such as Aib at position 2 or a 2'-O-methyl base. They come from the modification features of GenBank's patent sequences, EMBL-EBI's US patent proteins and the US patent office's sequence listings, copied on one day.

The Modifications column on Alignments gives how many modifications the hit's records mark, then the commonest names, each with how many: "3: Aib x2, Bip x1". A modification that two records mark at the same position, in the same words, counts once. Point at the cell to see up to 20, with their positions and the records they come from. The hit's pane lists the first 50 with their positions, words and feature keys. A record that marks more than 100 keeps its first 100, and a + follows the count.

Modified only, under Modifications beside US status, keeps the hits whose records mark at least one. Modification words keeps the hits with a modification whose words or feature key hold every word you type, even inside a longer word, so methyl also finds 5-methylcytidine. Case does not count, and ’ and # count as ', since some records write 2’-O-methyl or 2#-O-methyl for 2'-O-methyl. The CSV export and the workbook follow both. A search reads the modifications as it runs and keeps them. A search from before the add-on ran shows none; use Run again. Hits of an attached database show none.

The Patents tab has a Modifications column too. It gives the count and names in the same form, from each publication's own records of its matching sequences, so two publications that hold one sequence can differ. Point at the cell to see up to 20, each with its SEQ ID NO and position, such as SEQ ID NO 3: Aib at 2. On Patents, Modified only and Modification words look only at each publication's own records, so a publication whose own record of a modified hit marks none is left out. The Patents CSV gives the names and how many of each, by SEQ ID NO, in modified_seq_nos.

A note that only says a residue may be any of several, such as "Xaa can be any naturally occurring amino acid" or "Lys or Asn", marks a position that varies, not a modification, and Sequences leaves it out. Sequences shows what the records say. It does not work out a modification from a structure, a chemical name or a sequence's letters.

Dates

Each sequence is dated by the earliest date known for it, across its patent publications, its database entries and its papers.

  • A patent publication is dated by its publication date, not by its filing or priority date, even where its row shows them.
  • With Published before set, a sequence first published on or after your date still passes when one of its US or WO publications was filed, or claims priority, before your date, where the record gives those dates, since such a publication can count as prior art from that earlier date.
  • A UniProt entry is dated by the day its current sequence version appeared.
  • An ENA record at version 1 is dated by the day it first appeared in ENA. A later version has no date of its own, because ENA does not say when its sequence changed.
  • A paper is dated by the day PubMed gives for it: the earlier of the day it went online and its issue date. A paper that PubMed dates only by month or by year counts as dated before the first day of that month or year. So with Published before June 15, 2015, a paper dated only June 2015, or only 2015, still passes.
  • Without PubMed's records on the server, a paper is dated by its year only, in the same way.
  • A paper that PubMed does not hold is dated by the day or month Crossref gives for its DOI, where the server has Crossref's dates: the earliest of its print, online and issued dates. A month counts as it does for PubMed. When the earliest date names only a year, or the paper's own record gives an earlier year, the paper keeps its year. The hits CSV marks such a date "(Crossref)" in its paper_dates column.
  • A core_nt hit that ENA also holds shows ENA's description and papers. At version 1 it is dated by the day it first appeared in ENA, or by an earlier paper where the server has PubMed's records. Other core_nt hits are undated, and their papers do not date them.
  • A sequence with no known date reads "Undated" and always passes the date filter.

On Patents, with Published before set, the list holds the publications from before your date, the US and WO publications published after it but filed or claiming priority before it, marked "Filed before" ("Filed before your date" in the pane), and the undated ones. The later publications of the same sequences fold under the list, in one line such as "3 later publications of these sequences are folded. 1 of them is a US or WO publication with no filing date on record." Such a publication may still have been filed before your date, so check its page. Show them brings the folded publications back, each marked "After date" ("On or after your date" in the pane), and Fold them again folds them. A family keeps its publications from before your date in the list, and its pane counts the ones folded. Without a date nothing folds, and a new date folds them again. Applicants folds the later publications too, with the same line under its list, and its counts leave them out, so an applicant whose publications all came after your date leaves the list until Show them.

Filters and sorting

The filters change the list, and the CSV export follows them.

FilterKeeps
Find a sequenceRows whose query, hit or description holds the words
Minimum identity (%)Alignments whose Identity is at least this
Count an X as a matchWith Minimum identity, alignments whose Identity, X as a match is at least that value. An exact or pattern search does not have it
Minimum query coverage (%)Alignments that cover at least this share of the query
Minimum Query Match (%)Alignments whose Query Match is at least this. It shows where the search gives Query Match: see Results
Minimum query identity (%)Alignments whose Query identity is at least this. It shows where the search gives Query identity
Minimum hit identity (%)Alignments, or exact matches, whose Hit identity is at least this. It shows where the search gives Hit identity
QueryThe results of one query
Published on or afterSequences first published on or after the date, with no US or WO publication filed or claiming priority before it. With Published before, the sequences from one date up to the other
Published beforeSequences first published before the date, or with a US or WO publication filed or claiming priority before it. On Patents and Applicants their later publications fold under the list
Patent officePublications of one office
SpeciesHits with a record in one group, such as Mouse, or in one species. Each choice counts the hits it holds, before the other filters. On Patents and Applicants, the publications whose own record is in it
Drugs onlyHits that name a drug, on a search that names drugs. See Drug names
Claimed onlyHits whose SEQ ID NO a claim of one of their US publications names, on a search that reads claims. On Patents, the publications whose own claims name a hit's SEQ ID NO. See Claimed sequences
US statusHits with at least one US application whose status is in one group, such as Granted or Abandoned, on a search that reads patent status. On Patents, the US publications whose own application is in it. See Patent status
ModificationsHits whose records mark at least one modified residue or base, on a search that reads modifications. On Patents, the publications whose own records mark one. See Modifications
Modification wordsHits with a modification whose words or feature key hold every word you type, such as Aib. On Patents, the publications whose own records mark one with those words
Hit strandDNA hits on the plus strand or the minus strand
Patent wordsHits one of whose US publications uses every word in the text the server keeps for it. On Patents, the publications that use them themselves. See below
An area of Identity against coverageAlignments inside the area you drag across the chart, its identity and query coverage from and to. See Charts

Published on or after reads the dates that Published before reads (see Dates): a sequence passes when none of them falls before your date, and a sequence with no date passes both filters. A paper dated only by month or year counts here as published at the end of that month or year, so with Published on or after June 15, 2015, a paper dated only June 2015, or only 2015, still passes, unless another record of its sequence is dated before June 15. With both dates set, the lists hold what came between them, such as the intervening art between a claim's priority date and its filing date, and on Patents and Applicants the later publications of those sequences fold under the list as they do with Published before alone. Published on or after on its own folds nothing.

Patent words reads only the text the server keeps for What the patent says: the abstract, the first claim, the claims that name a SEQ ID NO, up to three paragraphs for each SEQ ID NO, and the sequence listing's notes. It does not read the rest of the description and claims, or publications of other offices, so a publication left out may still use your words. Every word must be in the same publication, in any order. Case does not matter, and a word matches whole words: antibody does not find "antibodies", but antibod* finds both. Words joined by hyphens match with or without them, so IL-2 finds "IL2" and not "IL-21". The filter needs a server with patent context and works on a search of up to 50,000 US publications. Its first use on a large search takes longer, while the server reads the search's publications.

Sort patents orders the rows of Patents, families or publications, by the highest Query Match or identity, the earliest or latest date, or the most sequences. Rows that tie go earliest first, undated last, and rows of one date go highest Query Match first. Sort matches orders alignments by bit score, E-value, Query Match, identity, identity with an X as a match, query identity, hit identity or query coverage, and alignments that tie go by their sequence's first published date, oldest first. A Within k changes search has no Sort matches: its rows go fewest changes first.

Saved views

Views, beside Columns above the list, keeps the way you read a result under a name: the tab, the filters, Sort patents, Sort matches, Group and the columns of each list. Type a name under Save this view as and choose Save view. The view is kept with the search, so anyone who can open the search can use it, and saving under a name the search already has updates that view. Saved views lists this search's views, then the views of your other searches. Use view sets the result as the view was saved. On another search it uses what fits, and one line says what it left out, such as a date on a search with no dates or the Patents tab on a search with no patents. A view's columns show for that result only: your own choice under Columns stays as it was until you change a column. Delete removes one of this search's views. A search keeps up to 50 views, each with a name of up to 80 characters. A combined search, an antibody CDR search and a sample result have no views. To try it, sign in and open any finished search, such as the two that Use demo data starts under Compare searches.

Exports and search records

Download for Lab Notebook saves a .labitem.zip for a completed paged search. In the Lab Notebook, open Inbox, choose Import item, then add it to a draft entry. Sign the entry to preserve the result with it. The package includes the search record and every stored matches row, including excluded matches. The page's filters and the Format choice do not change it. The search record states any result limits; this package does not recover hits the original search never kept. It is limited to 32 MiB expanded. For a larger result, download the record and CSV separately and attach them to an entry. It contains no live download links and needs no connection to Sequences when imported. It is a result snapshot, not a full search directory for restoring a job. The browser notebook demo keeps changes only until the tab closes; use a notebook archive to keep a copy. Combined searches and locally opened full records use their existing record and CSV downloads.

  • Export matches (CSV) saves the rows of the tab you are looking at that pass its filters, with their marks. On Patents it saves one row for each publication, whatever Group says, the folded later publications too, with application_number, filed and priority, then its patent family and its place against your date. A family is the publications that share a priority claim or an application, or, where no priority claim is on record, the same title and first applicant: family names its earliest publication, family_size counts its publications, and family_basis says how they were joined (source for a family that Patent families gives, priority, application, title, or single for a publication on its own). date_group is before, filedBefore (published on or after your date, but filed or claiming priority before it), later or undated, and empty without a date. why_listed is the line that opens the publication's pane. Then, where the server runs Claimed sequences, claimed_seq_nos gives the hits' SEQ ID NOs that the publication's claims name, each with those claims, and not_claimed_seq_nos the ones no claim names. Where the server runs Patent status, patent_status gives the status of the publication's application in the US patent office's words, patent_status_date the office's status date, patent_status_group its group, and patent_status_as_of the day the statuses were copied. Then, where the server runs Patent families, family_id gives the family's number at Google Patents Public Data, family_publications how many publications from 2001 on it lists in the family, and family_offices their offices; all three are empty for a family the rules formed. Last, where the server runs Modifications, modified_seq_nos lists each SEQ ID NO whose records mark a modification, with the names and how many of each, as "3 (Aib x2, Bip x1); 9 (Hyp x1)". On Applicants it saves one row for each applicant, with its other spellings, publications, joins and ROR suggestions. While the later publications are folded, it leaves them out, as the tab does. On Alignments it saves one row for each alignment, with the columns entries, papers, pubmed_ids, dois, paper_dates and paper_marks, and then the earliest publication of the hit's sequence, the one its title names: first_publication, first_kind, first_applicant, first_filed, first_application_number and first_priority. The filing and priority columns are filled only where a record gives them, as on Patents. A priority cell lists every claim, earliest first. Then come x_letters, the X or N letters of the stored sequence that face a query letter, and identity_x_as_match, which are empty for a pattern search. The columns species, taxa and species_groups list the organisms the hit's records name, their NCBI taxa and their groups. Then drugs and drug_sources name the drugs whose chains are the same sequence and credit their sources and license, where the server runs Drug names. Then come genes, gene_ids and gene_links: each gene's symbol, its NCBI Gene ID and how the hit links to it, in the same order, and empty for a hit with no gene (see Genes). Then claimed and claimed_in say whether claims of the hit's US publications name its SEQ ID NO (yes, no, or empty when none were read) and list each publication and SEQ ID NO they name, with the claims, where the server runs Claimed sequences. hit_identity is Hit identity (see Results). The next three, where the server runs Patent status, are patent_status, each US publication of the hit with its application's status in the office's words and the status date; patent_status_group, the groups as the US status column gives them; and patent_status_as_of, the day the statuses were copied. Then stored_codes counts the codes in the match of an Exact text or Pattern search with stored codes, and is empty for any other search. Last, where the server runs Modifications, modifications gives the hit's modifications as the Modifications column gives them, and modification_positions lists each one that Sequences keeps, up to 100 for each record, as "Aib at 2", separated by | . On CDR matches it saves the table on the screen.
  • Format, beside Export matches (CSV), shows when your administrator has set up export formats for the tab's file: on Alignments, on Patents, and on a combined search's references. Built-in (CSV) saves the file described above. Another format saves the columns your administrator chose, under their headers, as CSV, TSV or JSON lines, and can leave out or mask fields. A format marked external also leaves out every row that a confidential set found. Once the file is saved, a message says how many rows it holds and what it left out.
  • Export all tabs (Excel), beside it, saves the whole result as one Excel workbook, with a sheet for each tab in the page's order: Patents, Families when Group shows families, Applicants, and Alignments (Matches for a text or pattern search), then Aligned letters for a search that aligns, then a Search record sheet. Each sheet holds the rows of its tab's CSV, with the filters and sorts on the screen, under a header row that stays in view and can filter. Past Excel's limits on the rows of a sheet and the characters of a cell, the workbook cuts, and its Search record sheet says what it cut; export that tab as CSV for every row. A combined search has no workbook. Each sample result has Download as Excel, which works before you sign in.
  • Aligned letters holds the first 10,000 rows of Alignments, in the same order, each with its query, hit and positions, then the letters as the row's pane draws them, in three columns: query_letters, match_line, with | where the two letters are the same, and subject_letters. A gap is -. Where Align exactly aligned the row, the last columns give the exact alignment the same way. A cell holds up to 32,767 letters, so a longer alignment is cut there in all three columns, and Search record counts the cut cells; Alignments of selected rows (text) saves it whole. Filter the list to export the letters of other rows.
  • Sequences notes each export of matches, patents or a combined search's references, and each workbook: the format and its version, the file's SHA-256, its rows and what it left out, never what the file holds. Programs read a search's exports at GET /api/jobs/{id}/exports.
  • Download search record saves the search's record as JSON: the search engine version and each search program's version, the set, snapshot and additions it searched, every setting, each query, the PubMed data its papers came from, if any, and the SHA-256 of each file the search wrote. The engine version is the Search engine number of the release notes. For a search of a collection you imported, the button reads Download full search record, and the file holds every hit as well.
  • The Search record tab shows the main parts of the same record on the page, with its download button. It has no Export matches (CSV), which saves the rows of a list tab.
  • Export portfolio, on a row of Applicants, saves one applicant's publications with their sequences. See Portfolio export.
  • Download SVG and Download PNG, under each of the charts, save the chart as it reads now, with its title, its summary and its legend, on white. The SVG is 760 px wide and the PNG twice that, for a report or a slide.
  • More downloads, in the bar above the list on Alignments and Patents, saves three more files. Download FASTA saves the whole stored sequence of each hit once, named by its accession and its earliest publication, with the query it matched, its identity and its query coverage. FASTA of chooses the hits: The hits in the list, with the list's filters and marks, every page; Kept hits; or Selected rows. Alignments of selected rows (text) saves each selected row's alignment as C copies it, in the list's order. Kept patents as references (text) numbers each publication you kept, earliest first, with its number and kind, date, title and applicant. You select rows with their Select boxes (see Align exactly), up to 1,000 at a time. Sequences does not note these three files among the search's exports.
  • In the CSV and TSV files Sequences saves, a text cell that starts with =, +, -, @, a tab or a carriage return begins with an apostrophe, so a spreadsheet shows it as text and never runs it as a formula. A number, even a negative one such as -1.5, is saved as it is. The workbook from Export all tabs (Excel) keeps such text as text, with no apostrophe. Sequences takes the apostrophe off again when you load one of these CSV files as a result file.

The results page and the search record state the search's limits, such as how many sequences each query returns, and end with "These results do not establish patentability."

Run again

Run again starts a new search with the same request as a finished one: the same queries, settings, snapshot and additions. It finds an older snapshot by its ID while that snapshot is still on the server. Once the snapshot is gone, it answers "Snapshot not found." An antibody search runs again on the CDRs it found, without numbering the chains again. A search that a case's plan started has no Run again: open the case and run its plan again.

Run again sits in a finished search's result, and on the card and in the status bar of a search that failed, was cancelled or was interrupted. A search that still waits or runs cannot be run again: "This search is still waiting. Run it again once it has ended, or cancel it first." While the search a Run again started waits or runs, another Run again starts none. A program that asks gets 409 with the code not-allowed-in-state.

Watches

A watch follows a finished search of the snapshot, or an applicant, through each addition of newer records to the snapshot. It shows only what is new: matches the search has not found before, or the applicant's new publications.

  1. To watch a search, open a finished search of the snapshot, choose Watch this search, then Watch. The button then reads Watching. To list what the search holds already from a day on, such as everything since the last office action, set Count as new from before you choose Watch: the matches dated on or after that day show as new under Your watches at once, with no email, and the other matches count as seen. A patent record is dated by its publication, and a database entry by its own date. A match with no date counts as seen.
  2. To watch an applicant, choose Watch this applicant on its row under Applicants, or write its name under Watch an applicant in Your watches, below Recent jobs on the search page, and choose Watch. The form watches the applicant in the newest snapshot that has patent sets. Case, accents, punctuation and endings such as Inc do not matter, and the names you have joined to it count too. The form does not look the name up, so a watch on a name that no publication names first lists nothing until a new publication does. A watched row reads Watching, which goes to the watch.
  3. Your watches lists each watch with its state: Watching, Running, Paused or Needs attention. A watch on a search opens its results. A watch on an applicant names the applicant and how many names are joined to it, and has Export portfolio. Each watch can be paused, resumed or deleted. A watch paused for its size can only be deleted, since it would pause again at the next addition.
  4. The server looks for additions every hour, so a watch runs within about an hour after one arrives. A run that fails is tried up to 3 times.
  • A watch on an applicant runs no search. After each addition it lists the applicant's new publications, each with its date and number of sequences, linked to the patent.
  • As under Applicants, a publication counts for the first applicant it names. A watch on an applicant finds only what the patent sets hold, US published applications among them. See Sets.
  • Needs attention means the watch has a problem or its last run did not finish. The line under it says why, and whether the watch tries again within the hour or waits for the next addition.
  • A watch whose snapshot left the server says so, and the line goes within the hour once the snapshot is back.
  • An account may have 50 watches, of searches and applicants together.
  • A watched search may hold up to 50,000 matched records. A watch whose results grow past that pauses and says so: delete it, then watch a narrower search.
  • Antibody searches cannot be watched.
  • On the hosted service, once your free trial has ended, your watches do not run, as new searches do not start, and each says so. Write to support@conductscience.com to choose a plan.
  • What your watches find shows under Your watches, and the line under that heading says whether this server also sends watch emails. Each email goes to your address and says only how many new matches or publications one of your watches found, with a link to sign in and see them and a link that pauses the watch without signing in. It names no watch, query or match. A server's staff can choose emails with details instead, which name the watch and list what it found.

To try it, choose Use demo data beside Watch: it fills the applicant that US12788009B2 names first.

Portfolio export

A portfolio export saves one applicant's publications with their sequences, as a CSV file and a FASTA file.

  1. Choose Export portfolio on the applicant's row under Applicants, or on its watch. The export runs as a job in Recent jobs.
  2. It collects every publication in the snapshot and its additions whose first-named applicant is this one or a name you have joined to it, with each of their sequences in Patent DNA and Patent proteins.
  3. When it finishes, Download CSV and Download FASTA save its files, on the applicant's row under Applicants and on the export's card in Recent jobs. The row and the card say what the export holds, such as "The export holds all 2 sequences, from 1 publication." On the row, Export again starts a new export, for example after a new join.
  • The CSV has one row for each sequence, with the columns publication, published, applicant, seq_id_no, accession, molecule_type, length and sequence. The FASTA file holds the same sequences.
  • An export holds up to 100,000 sequences and 100 million letters. It takes the oldest publications first, and each publication's sequences by SEQ ID NO, so an export that stops at a limit leaves out the newest publications. Its line then says how many of the applicant's sequences it holds, from how many of the oldest publications.

To try it, watch the demo applicant (see Watches) and choose Export portfolio on its watch.

Cases: claims to search plan

A case turns a patent application's claims into a search plan, which you check and lock before any of its searches run. The case then ranks what the searches find against each claim, and you save its report as PDF or Word.

Log: every action in the case goes in its log, which Verify the log checks.

One claim, followed through the six steps:

  1. Claims: claim 1 recites "a protein at least 90% identical to SEQ ID NO: 1"
  2. Review: you accept SEQ ID NO: 1 at 90% identity
  3. Plan: a search of SEQ ID NO: 1 in the patent proteins and one in Swiss-Prot, each at 90% identity or more
  4. Run: the locked plan runs both searches
  5. References: each match, ranked against the claim, such as "Meets 90%: 97.3% identity over the full query"
  6. Report: the matches you keep, with their papers and alignments, and the sentences you accept on why they count

Start a case

Start a case under Claims to search plan. Only the Case name is needed:

FieldNeeded to startWhat it does
Case nameYesNames the case in your list
Sequence listingNoHolds the sequences the claims recite by SEQ ID NO. Import the listing under Sequence tools first. You can choose it here, or later in the Plan step
SnapshotNoThe snapshot to search: a read-only copy of public sequence files, built on a known date. See Sets
Application numberNoNames the application in the case header and the report
Prior art dateNoNew searches look for what was published before it: the plan drafts each search with Published before set to it (see Dates). Leave it empty and a publication XML file gives its filing date

The case header shows the application number and the prior art date, such as "Prior art date 2020-03-02 (filing date)". Choose Change to edit them. A locked plan version keeps the dates it was locked with.

Case summary, under the case header, opens to one paragraph about the whole case, filled in from a fixed template: see Summaries.

Try it with demo data. Import the demo listing of US12788009B2 (us12788009b2-listing.xml), then choose Use demo data beside Import a listing: it names a case for US12788009B2 and chooses that listing. In the case's Claims step, Use demo data fills the grant's 9 claims as text or as a text file (us12788009b2-claims.txt), or the grant's full text (us12788009b2.xml), which brings its application number and filing date. The case is a demo case: see Demo data.

Where a server has examples, the examples page at /samples links a sample case: a real published patent made into a case, searched and saved read-only. See Examples and sample results.

Step 1: Claims

Bring in the claims in one of three ways:

ChooseFor
Paste the claimsClaims as text
Publication XML fileOne publication from the US patent office's full-text files, an application or a grant, up to 20 MiB. The publication's number and dates come with its claims
Word or text fileClaims in a .docx or text file, up to 15 MiB
  • A weekly full-text file holds many publications one after another, so save the one you need as a file of its own.
  • When the file gives an earliest priority date too, from a foreign application or a US provisional application, the step offers Filing date, Earliest priority date or Another date as the prior art date. Use this date saves your choice.
  • Under What the description says about each claimed SEQ ID NO, the step gives a summary of each SEQ ID NO the claims recite. See Summaries.
  • A case can hold the claims as filed, as amended, as published and as granted: see Sets of claims.

Sets of claims

A case can hold an application's claims as filed, as amended on a date, as published and as granted. Each import is a set of its own, and the case keeps every set with its source, date, SHA-256 and who added it. These claims are says which set an import is:

SetWhere it comes from
As filedClaims you import and mark as filed
As amendedClaims you mark as amended, with their Date of the claims, or a Word file marked up as an amendment
As publishedAn application's publication XML, which says it for you, with the publication's date
As grantedA patent's publication XML, which says it for you, with the publication's date
  • A Word file marked up as an amendment reads as the claims as amended: struck-through and deleted text is left out, and underlined and inserted text is kept. In a claim marked "(Currently amended)", text in double brackets, such as [[about]], is left out too.
  • Pasted text loses strike-through, so for an amendment send the Word file. An older Word file (.doc), a Word file saved with a password, or a PDF cannot be read: save it as .docx, or paste the claims.
  • A new set becomes the current set, unless it is dated before the current set. Then it is kept beside it, and Make current, under Sets of claims, makes it current, with an optional note for the log.
  • What changed compares any two sets claim by claim. Each claim is unchanged, amended, added or cancelled, and a renumbered claim shows its number in both sets. In an amended claim, added words are underlined and removed words are struck through.

When the current set changes, an unlocked plan loses its searches. A locked plan stays locked and follows the set it was locked on.

What else follows a change of the current set
  • The rules find the new set's features again. Each decision you made carries over wherever a claim's words did not change, even when the claim was renumbered. No model is asked again.
  • A locked plan's searches run on the claims it was locked on. To search the current set, unlock the plan with a reason, which the log keeps, and draft it again: Unlock the plan asks for the reason first.
  • The report names the set the plan follows, with its label, date, source and SHA-256.

Step 2: Review

The case reads each claim and marks what it recites as features. Accept, reject or edit each feature, or select words in a claim to add one. The features you accept make the plan.

The case marks:

  • SEQ ID NOs and identity thresholds, as in "at least 90% identical to SEQ ID NO: 1"
  • Fragments, complements, fusions, regions and CDRs
  • Sequences written out in the claim
  • The changes a claim allows, such as "one, two or three amino acid substitutions"
  • Letters a claim reads as any residue, such as "wherein X is any amino acid"
  • The residues a claim allows at a position, such as "wherein X at position 26 is lysine (K) or arginine (R)"

Where the server offers suggestions from a language model, the step names the model, and a button marked "(Pilot)" asks it for features the rules missed. None counts until you accept it. Without a model, the step says "Model suggestions are off."

The step sends the model the claims and, for claims from a publication XML file, up to 40 paragraphs of the description that cite a SEQ ID NO the claims name or define a term they use. The server's administrator chooses the model and where it runs: see Claim suggestions from a model.

How model suggestions work
  • A model can take minutes: the step shows how long it has been asking, and you can keep working.
  • If a form is open when the model's suggestions arrive, it stays open with what you typed. A line says newer features are ready, and they show when you save or cancel the form.
  • The server asks for one case at a time. While the model answers for another case, the step says to ask again in a few minutes.
  • Each suggestion names the model, and its threshold, range or length counts only when the claim's own words state each number. A suggestion may quote the paragraph of the description it draws on.
When you decide a feature again
  • Accepting a feature already accepted, or rejecting one already rejected with the same note, changes nothing and adds nothing to the log.
  • A feature you accept after rejecting it drops the rejection's note, which the log keeps, and a model's suggestion shows the model's note again.

Step 3: Plan

The plan lists the searches your accepted features call for. Check it, change it if you need to, then lock it. Only a locked version runs.

  • A draft plans a search of each set on the server that holds the feature's type, patent sets first: for a protein, the patent proteins and Swiss-Prot.
  • Each SEQ ID NO is searched once, at the lowest identity any claim gives it.
  • Set to search lists every set of the search's type, from any snapshot on the server, and each attached database.
  • TrEMBL, and an attached database such as core_nt, take much longer, so they join only when you ask: a button such as Add TrEMBL adds a copy of each fitting search and says how long they should take.
  • Query groups shows how the plan groups its SEQ ID NOs, and Size and time what it takes to run: see Query groups and large plans.

Each search's minimum identity is the one its claims need, and its reason says why:

The claim givesThe search's minimum identityIts reason, for example
An identity, such as "at least 90% identical"That identityThe claim's identity
A SEQ ID NO with letters that stand for more than one residue, such as X for any amino acidLower, by what those letters take away"Keeps 70% identity or more, since identity counts 3 of its 15 letters, X and Z, as differences."
Changes, such as "with one, two or three amino acid substitutions"What that many changes of any kind leave"Keeps 80% identity or more, which 3 changes leave in its 15 letters."
Why a search's identity can be lower than its claim's

A search's identity counts a letter that stands for more than one residue as a difference, while References reads it as each residue it stands for. A claim that allows changes is met by a match with that many of those changes or fewer. So the search finds every match that could meet the claim, and References then reads which matches do.

Query groups and large plans

A draft groups the claimed SEQ ID NOs by molecule type, by length, by the identity their claims give them and by the method that searches them. Each group is one search on each set it covers. Lengths fall in three bands:

  • Short: DNA under 50 bases, or a protein of 30 residues or fewer
  • Up to 1,000 letters
  • Longer

Under Query groups, each group lists its members and the claims and features it serves, with a line such as "41 proteins of 80 to 450 residues at 90% identity or more: Smith-Waterman on Patent proteins and Swiss-Prot, 2 searches as 6 jobs." To see groups, draft the plan of the demo case. Its claims recite two SEQ ID NOs, too few to go over the limits below.

Change the groups before the lock
ChooseTo
Move SEQ ID NOsTake some of a group's SEQ ID NOs to another group of the same molecule type that searches the same sets with the same E-value, hit count and date
SplitPut some of a group's SEQ ID NOs in a new group with the same settings
Join another groupBring a group on the same sets, with the same engine, E-value, hit count and date, into this one

A group that takes SEQ ID NOs searches at the lower identity of the two, so each SEQ ID NO still gets what its claims need. The changed searches get new reasons, and the log records each change.

Size and time gives the plan's queries, letters, searches and search jobs, and an expected time from this server's timings. It names each of the server's limits the plan goes over, and what the plan does about it.

The server's limits, and what a plan does about them
LimitWhat the plan does
A search job takes up to 20 queries and 100,000 lettersA search of many SEQ ID NOs runs as several jobs
A search may run for one hour, unless the server sets another limitA job that would run past it holds fewer queries. A query that would go over it even alone is named, with a faster engine, or part of the sequence, as the way to search it
A run holds up to 100 search jobsA larger plan runs as several runs, one after another. The step lists them in order, each with its searches, jobs and expected time

Lock the plan

Lock the plan when it is ready. Each lock makes a version:

  1. Draft: the plan you check and change
  2. Lock as version 1: records the plan and its SHA-256
  3. Unlock the plan: under Unlock to change it
  4. Lock again: makes version 2, and so on

Each locked version stays, and the Run step can still run an earlier one. If the server stops while it locks the plan, the plan stays unlocked and no version is made: lock it again.

From a locked version, Download search request (Word) and Download search request (JSON) save what a search team or another system needs to run the plan: the case, the application number and the prior art date, then each group with its sequences in FASTA, its method, sets, thresholds and reasons, and the claims and features it serves. Each file ends with the plan version and the SHA-256 its lock recorded. The log records each download with the file's own SHA-256.

Step 4: Run

Pick a locked version. The case checks that it has not changed since you locked it, then runs it. Each run shows a line such as "7 of 12 search jobs done, about 4 minutes left". The results open with the plan's filters.

  • A large plan, such as one search of 250 SEQ ID NOs, needs more jobs than the queue holds. The case starts what the queue has room for and starts the rest as jobs finish. It always leaves 2 places for searches from the form.
  • A plan of more than 100 search jobs runs as several runs, one after another. Each starts on its own when every job of the run before it has finished, and its title says which it is, such as "Plan version 2, run 1 of 3".
  • A search of many queries runs as several jobs, and the time left counts how many jobs the server runs at once, with any servers that share its queue.
  • When the server restarts, the run goes on by itself. A search that failed has Run this one again.

Step 5: References

References lists every publication and database entry the case's searches found, under the claim feature whose search found it, ranked by fixed rules. Here a feature is one or more SEQ ID NOs you accepted in Review, with the identity, fragment length and allowed changes that apply to them, or a sequence written out in a claim. Each feature shows the claim's words, such as "at least 90% identical to SEQ ID NO: 1", its threshold and the searches that serve it.

Which searches serve a feature
  • A search the plan ran serves the features its plan search covers, and keeps the filters it was locked with.
  • A search you add in Report serves the features that recite its query's SEQ ID NO when it ran on the case's listing. So does a result file you import whose queries are named SEQ_ID_NO_<n>, as the search request's FASTA names them.
  • The references of any other search are under Other searches, which has no threshold.

Summary of the results opens the step once a search adds references, with what the case's searches returned before your marks: see Summaries.

By feature lists the references under each feature, as below. All references lists each reference once across the case: see All references.

Inside a feature, the references rank by four rules in order: each rule breaks the ties that the rules before it leave. No language model takes part, so the same results always give the same order.

  1. Meets: a best match that meets the claim comes first
  2. Coverage: more of the claimed region covered
  3. Date: before the prior art date, then a US or WO publication filed or claiming priority before it, then undated, then on or after it
  4. Query Match, then E-value: the higher Query Match, then the lower E-value

Ties go to the earliest publication, then the ID. With no prior art date, dates do not rank.

A claim with no identity reads as written: identical over the whole query. The Meets column reads what the best match does against the claim:

Meets readsThe best match
MeetsMeets the claim's threshold
BelowIs below the claim's identity
More changesHas more changes than the claim allows
Other changesHas changes of other kinds than the claim allows
DiffersDiffers from a claim that reads as written
How a match meets a claim
  • With a fragment length, the identity counts over an alignment at least that long.
  • A letter in the query that stands for more than one residue matches each residue it stands for, as WIPO ST.26 defines it: in a protein, X for any amino acid, B for D or N, Z for E or Q, and J for I or L; in DNA or RNA, N for any nucleotide and the other IUPAC codes, such as R for A or G.
  • A letter the claim reads as any residue, as in "wherein Z is any amino acid", matches any residue, even where ST.26 gives the letter a narrower meaning.
  • A claim that allows changes, such as up to 3, is met by a match with that many changes or fewer of the kinds the claim names, and no change of another kind: "up to 3 substitutions" is not met by a deletion. A claim that names no kind, such as "up to 3 mutations", counts substitutions, deletions and additions alike. Query letters outside the alignment count as deletions.
  • A claim that gives an identity as well as changes is met by a match that meets either one.
  • Sequences does not check whether a substitution is conservative, and says so.

Each row shows its rank and each score beside it: Meets, Coverage, Date, Query Match, E-value and the mark. A narrow screen leaves some of these columns out. Open the row to see them all. A percentage that misses a threshold is never rounded up to it: 89.96% shows as 89.96%, not 90.0%.

Open a row: its keys, and what it says

Click a row, or press Enter on its name, to open it. Escape closes the row, and the Up and Down arrow keys move between rows.

A publication's open row starts with the line that says why it is listed, as on Patents (see By patent). It names the query as the claims' SEQ ID NO, as in "is identical to SEQ ID NO 1 of the claims over all 31 letters", says where the publication falls against the prior art date, and counts the other publications of the case's results that hold the query unchanged, with the earliest of them.

Then the open row gives the scores in words, the search and row of the best match, and Open results, which opens that search's results:

The open row saysFor
"Meets 90%: 97.3% identity over the full query"A match that meets a 90% claim
"Meets as written: 93.3% identity over the full query, and 1 position is X, read as any amino acid"A claim that reads as written, with an X in its sequence
"Meets up to 3 substitutions: 1 substitution, 93.3% identity over the full query"A claim that allows changes: the words count the best match's changes of each kind
"Changes other than substitutions: 2 deletions, 86.7% identity over the full query"A match with a kind of change the claim does not allow
"2 substitutions, not checked as conservative"A claim of conservative substitutions
"Published 2011-09-22, before the prior art date"The reference's date against the prior art date

Marks come from each search's results: to keep or exclude a reference, mark its row Keep or Exclude in a search that found it (see Keep and exclude). Open results, in the reference's open row, opens the search of its best match. The Mark column reads "Kept", "Excluded" or "Mixed", and the open row names each search's mark, such as "Kept in search 1, excluded in search 3". A reference that every search that found it excluded leaves the list. Show excluded lists those references.

More rows and downloads
  • For a feature with more than one SEQ ID NO, a row says how many it matches, such as "Matches 37 of 250 SEQ ID NOs".
  • Each feature lists 20 references at first, and a button such as Show 20 more adds the next ones.
  • Download CSV saves every reference of every feature, with the claim's words, the scores and the best match. The excluded references come last in their feature, with no rank. Format, beside it, offers your administrator's formats for a case's references, as on a result.

Under a feature, Follow-up searches suggests searches from fixed rules, each with its rule and why:

It suggestsWhen
A more sensitive methodNo reference meets the claim
The part of the claimed sequence that the best reference leaves outThe best reference covers under 80% of it
A set the server holds that no search of the feature read, from whichever snapshot holds itSuch as Swiss-Prot beside a snapshot of the patent sets alone

Add to plan puts one in the draft plan with the rule as its reason, and Go to Plan opens the plan. A locked plan unlocks and takes it in its next version, and the log names the follow-up search as the unlock's reason. Dismiss hides one.

Reading the list changes nothing in the case but its log, which records each follow-up search the first time it shows. Each download adds a line to the case log with its format, its rows and the file's SHA-256.

All references

All references, beside By feature in the References step, gives each reference one row, however many features list it and however many searches found it. Rows rank by the best rank a reference reaches under any feature, then by how many claim features it meets, then the earliest publication, then the ID. No language model takes part.

ColumnShows
RankThe reference's place in the list
ReferenceThe publication or database entry
MeetsThe features whose threshold the reference meets, such as "F1, F3", or "None"
Best rankIts best rank and the features it reaches it under, such as "1 under F2 and F3"
DateShown when the case has a prior art date
FamilyThe family's size, when it holds more than one publication
MarkThe mark when every feature marks the reference alike, and "By feature" when they differ
  • A narrow screen leaves some columns out. Open the row to see them all, and what the reference is under each feature: its rank there, how its best match meets the claim, and its mark, such as "F2, claim 1: rank 1. Meets as written: 100.0% identity over the full query. Kept." Open results opens the search of its best rank.
  • A reference excluded under every feature that lists it leaves the list. Show excluded lists those references.
  • Download CSV saves the list, each reference once, with its rank, the features it meets, and its rank, best match and mark under each feature. Format offers your administrator's formats, as for the list by feature.

Step 6: Report

The report puts the case in one file, with the publications and matches you keep and the sentences and summaries you accept or edit. Download PDF and Download Word save it. The server keeps no copy, and the log records each report's SHA-256.

The report givesFrom
The case's application number and prior art dateThe case header
A table of the searches of the latest locked plan version, with each one's query, method, set, thresholds and reason, and then the features no search coveredPlan
For each search, its settings, the publications and matches you kept, with each kept match's papers and alignment, its top 15 publications by best Query Match and its top 25 matchesThe plan's searches, and up to 100 other searches that you add under Add a finished search or bring in under Import a result file (see Result files from other programs)
By applicant, after the table of searches: the publications you kept, and the publications that hold each match you kept, by applicant across all of the case's searches, with the applicant names you joined (see By applicant)Up to 100 applicants, each with up to 20 publications, and it says how many more there are
Your notesNotes: up to 500 in a case, each up to 5,000 characters
The claim charts, with your answers and notesClaim charts: see Claim charts
The sentences you accept or editWhy each reference counts: see Why each reference counts
The summaries you accept or editSee Summaries

Its publication tables add a Filed column when a publication's record gives its filing date or priority. Download decisions (CSV) and Download decisions (JSON) save every decision in the case: see The decisions file.

Claim charts

In the Report step, Claim charts sets each claim's features against its top references, one table a claim. Each row is a feature of the claim, and each column a reference. Until you choose, the independent claims are charted, each with its 5 top references: those you keep first, then by rank in the References step.

A row forIts cell says
A SEQ ID NO or a written-out sequence, with the identity, changes, length and region that qualify itMet when the reference's best match meets the claim, Not met when it does not, as the References step decides. The facts: the identity, where the match aligns, and the reference's own SEQ ID NO
A choice of residues at a positionThe residue the best match has there, from its alignment: Met when the claim allows it, Not met when it does not
A complementMet when the match is on the minus strand, the complement of the query
A length with a maximumMet when the reference's stored sequence is within it
Anything else, such as a fusion or CDRsNot shown, with the reason: the rules do not read it

A cell with no match to read says Not shown. Fixed rules fill in every cell, so the same case always gives the same chart. Each fact names its source: the result row opens the search's results, and the reference's number opens its page.

  • Set answers a cell yourself: Met, Not met or Unclear, with a note. The rule's answer gives it back. A cell you set says who set it and what the rule says, and says so when the rule's answer has changed since.
  • Choose the claims to chart picks up to 20 claims. A dependent claim with no SEQ ID NO of its own reads the matches of the claim it depends on.
  • Add a reference and Remove change a chart's references, up to 10. Use the top 5 goes back.
  • Download CSV saves every cell, one line each, with its facts, sources and your answer. Format offers your administrator's formats, as for the references.
  • The report prints a table for each claim and reference: each feature, its answer and what the match shows.
  • The case log records every choice and answer, and the decisions file lists the answers.

Why each reference counts

In the Report step, Why each reference counts lists each search in the case. For each publication you keep, and for the 5 publications of each search with the best Query Match, the case drafts up to four sentences from what it stored. Each sentence states one fact:

SentenceStates
AlignmentThe best alignment: which sequence of the publication matches which query or claimed sequence, its Identity as identical residues out of the alignment's length, and its Query Match
PlanThe identity a claim recites, which the plan's search used as its Minimum identity (%). Where the query's letters that stand for more than one residue lowered the search's identity, the sentence gives both and says why. Where a claim's allowed changes set it, the sentence names the claim and the kinds of change it allows, and the identity they leave. For conservative substitutions it adds that they are not checked as conservative
DateThe publication date against the case's prior art date: before it, or on or after it. The sentence names the date: the filing date, the earliest priority date or the prior art date
PaperThe genes and diseases PubTator 3 names in the abstract of the earliest paper on the sequence that has any, where the server has PubTator 3's data

A sentence that has no fact to state is left out. Each sentence names its source, with a button or link to it: the result row, the plan version, or the publication's or paper's page. The templates write the sentences, so the same facts always give the same words, unless you accept a model's draft in their place (Drafts from a model).

  • Choose Accept, Edit or Reject for each sentence. An edit keeps the draft beside your words. A sentence you neither accept nor edit stays out of the report.
  • When the facts behind a sentence change after you decide, for example when you reject a feature that the sentence names, the sentence says that its draft changed, and it leaves the report until you decide again.
  • The report prints the accepted and edited sentences of the publications you keep, with their sources. Top publications you have not kept are listed in a collapsed section, such as "2 top publications not kept". What you accept there goes in the report once you keep the publication.
  • The case log records every decision.
Which date, and facts that change while you read
  • The date is the case's prior art date. A case with none uses the filing date that came with a publication XML file, and for a search the plan ran, the search's Published before date. With no date, there is no date sentence.
  • When the facts behind a sentence change between the time the page shows it and your click, the click decides nothing. The page says "This sentence changed after you read it, because its facts changed. Read it again before you decide." and shows the sentence as it reads now.

Summaries

Sequences writes short summaries in a case from fixed templates and the case's own facts, so the same facts always give the same words in the same order. Where the server names a model, you can also ask it for a draft of the case summary, the summary of the results and each claimed SEQ ID NO's summary (Pilot): see Drafts from a model.

SummaryWhereSays
The caseCase summary, under the case headerThe application and prior art date, the claims and their sets, the features by kind and how many wait for review, the listing and its WIPO ST.26 check, the claimed SEQ ID NOs, the plan and its searches, then where the case stands, such as "Plan version 1 ran: 2 of 2 search jobs done."
The resultsReferences, first in the stepHow many references the case's searches found, each once, how many meet a claim feature's threshold and came before the prior art date, the first 5 in rank order, whether each claim feature has a reference that meets it, and the searches they come from
A claimed SEQ ID NOClaims, under What the description says about each claimed SEQ ID NOThe claims that recite it and what they recite, then the paragraphs of the description that cite it, quoting the first sentence of each that does while the summary has room
A referenceReferences, in the open row, after Open resultsIts feature, its dates, applicant and title, its own SEQ ID NOs that matched, its best match, the scores that rank it, the prior art date and the searches that found it
A searchReferences, after the featuresWhere it came from, its method, the data and filters it used, its queries and hits, and what it found for each feature, with the first references in rank order
  • Choose Accept, Edit or Reject. An edit is one paragraph in your own words, and the summary keeps the draft beside them. A summary you neither accept nor edit stays out of the report. A line in each step counts the summaries decided and the ones in the report.
  • When the facts behind a summary change after you decide, for example when you change the prior art date, the summary asks you to decide again, and it leaves the report until you do. The summary of a reference that you exclude after deciding also leaves the report.
  • The report prints the summaries you accepted or edited, each with the template that filled it in and who accepted it, and when. The case summary comes first, before the case's details. The rest are under Summaries, after By applicant, the summary of the results first. The case log records every decision.
  • Pasted claims come without a description, so each SEQ ID NO's summary names only the claims. Import the publication's XML file, and the summaries quote the paragraphs that cite each SEQ ID NO.
How summaries are filled in and checked
  • There is one summary of the case, one of the results once a search adds references, one for each search that adds references, one for each feature's first 5 references and the ones you keep, and one for each claimed SEQ ID NO. Claims lists the first 10 SEQ ID NOs, and a button such as Show 10 more adds the next ones, up to 200.
  • A search's summary and the summary of the results describe what the searches returned, before your marks, so keeping or excluding a reference never changes them. A line over each says so. Your marks show in the Mark column of the references, and in the report in each search's own section.
  • A summary is filled in again from the case each time you open it, so it says what the case holds now. Each sentence names its source: a claim, a paragraph of the description, a search, a result row, a publication or the case. Sequences checks that every number, date, SEQ ID NO and publication number in a summary is one its facts hold, and that every quote is in the paragraph it names.
  • Under each summary, "Filled in from template" names the template and the start of its SHA-256, so you can tell which wording filled it in. A change to a template's words gives it a new name.

Drafts from a model

Where the server names a model, four texts have a button that reads "Draft with" and the model's name, marked "(Pilot)": the case summary, the summary of the results, each claimed SEQ ID NO's summary, and the sentences on each kept publication in the Report step. A search's summary and a reference's summary have no such button. The model gets that text's facts, each with its source, and nothing else.

  • Its draft shows under the template's text with the model's name, how long it took and its tokens: accept it, edit it or reject it.
  • A draft is dropped when a quote in it is not in its source or a number in it is not in the facts. The box then says which check failed, and the template's text stays.
  • Only a draft you accept or edit reaches the report, marked "(Pilot)", and the case log keeps each draft and decision.
  • The model answers one request at a time for the whole server, drafts and claim suggestions alike. A draft can take minutes, and you can keep working while it drafts. While it answers another request, the step says to ask again in a few minutes.

Result files from other programs

You can run a search outside Sequences, such as BLAST+ or FASTA36 on your own computer against the patent sequences, and bring its results into a case. Results of the US patent office's sequence search come in the same way, as office actions print them, so an examiner's searches can sit beside Sequences' own. The import becomes one of the case's searches. It opens in the same results view as any search, with marks, families, dates and the sentences under Why each reference counts.

  1. Choose the file: under Import a result file, in Report
  2. Match the columns: if the header is new to Sequences
  3. Import the file: it becomes one of the case's searches
  4. References: lists its references with the others

Sequences reads these files, as they are or gzip compressed:

FileWritten by
BLAST tabular-outfmt 6, with its 12 standard columns, or -outfmt 7, whose # Fields: line names the columns
BLAST XML-outfmt 5
BLAST JSON-outfmt 15
FASTA36 tabular-m 8, with or without its comment lines
US patent office search resultsThe US patent office's sequence search, as office actions print it: protein and DNA searches, among them those that translate DNA
CSVAny program, with a header row and then one row for each match. A CSV file that Sequences saves can come back this way
Comma- or tab-separated textAny program, with a header row whose columns you match before the file uploads

On the command line, blastn runs megablast unless you add -task blastn. Megablast finds close matches and can miss weaker ones. Sequences runs blastn, so add -task blastn when a file should compare like for like with Sequences' own search.

A file with a header row whose names Sequences does not know opens Match the columns before it uploads. A CSV file whose names it knows imports as it is, and Match the columns by hand opens the same step for it.

The column names Sequences knows

The names in a CSV header match in any case: qseqid or query, sseqid or subject, pident or identity, length, qstart, qend, sstart, send, evalue, bitscore or score, publication and seq_id_no. Each row needs a subject, or a publication and a SEQ ID NO.

Match the columns
  • Sequences names the delimiter it found, comma or tab, and the line of the header row. When a title or notes come before the header, choose its line under Header row.
  • Under each field, choose the column that holds it: Query, Hit, Publication, SEQ ID NO, Identity, Query coverage and Score. A row joins by its hit, or by its publication and SEQ ID NO, so match the hit or both of those. A column fills one field.
  • The table under the fields lists each column with the field it fills and its values in the first rows. A row with more or fewer cells than the header is skipped, and the line above the table names it.
  • Every other column is kept with its row. In the results, a row lists them under From the file, with its line in the file. These results sort by identity.
  • Identity and coverage read as percents. When every identity is 1 or less, the import report says so: multiply the column by 100 and import the file again.
  • The case keeps the map with the import, and the next file with those columns is offered it again.
  • After the import, Import report under the search says which lines Sequences read and which it skipped, and why, such as a line with fewer cells than the header or the header row again. The case report gives the same, with the map.
  • A file can have up to 1,000 columns.

Try it with demo data. Choose Use demo data under Import a result file. It chooses us12788009b2-blastp.tsv: BLAST+ tabular output for the 16 peptides of US12788009B2 against s1-patent-proteins.fasta. Its rows join where the case's snapshot holds those grants; the rest stay in the results with the reason.

Use a demo CSV, beside it, chooses blastp-demo.csv instead: BLAST+ output for three published proteins, saved as CSV with BLAST+'s own long column names, such as % identity. A CSV with those names opens Match the columns, with each column matched by its name. On a server whose patent sets hold its accessions, 12 of its 14 rows join, and the other 2 name a Swiss-Prot protein that no patent holds.

How a row joins a stored sequence
  • Each row joins a stored patent sequence of the case's snapshot and its additions: by accession, such as AX123456.1 or gb|AX123456.1|, or by publication and SEQ ID NO. A publication number can be written as US20200012345A1 or US-2020-0012345-A1, or with spaces and a slash. A publication written without its kind code, such as A1 or B2, joins only when one kind holds that SEQ ID NO.
  • A search of a Sequences snapshot's own files, such as a copy of the patent sets on your own computer, names each sequence by its place, such as cb26. Such a row joins when the file names the database by a path that includes the snapshot's folder and the case searches that snapshot. BLAST+ writes the database in -outfmt 5, 7 and 15, and FASTA36 in -m 8 with its comment lines. A plain tabular file names no database, so its places do not join, and the reason says so.
  • A file that names its program joins the set that program searches, such as Patent DNA for blastn and Patent proteins for blastp. A CSV file, or a file that names no program, joins both. Look up its sequences in picks one set.
  • A row shows its alignment when the file has one. BLAST XML and JSON files have alignments, and tabular files have them when they include the qseq and sseq columns.
What Sequences reads from US patent office search results
  • Each hit, in rank order. Its summary row gives its Query Match, score, length and database part. Its RESULT block, when the file has one, adds its record, its identity and its alignment. A hit with no RESULT block keeps what its summary row gives.
  • A hit joins the patent sequence its record names: by publication and SEQ ID NO (Patent No. or Publication No. with Sequence 5, or a record's PN and PS lines), or by its accession. A sequence named only by its application, such as US-08-940-968-24, cannot be looked up, and the reason says so.
  • The file's search says what it searched, so its hits are looked up in Patent proteins or Patent DNA.
  • A search that translates DNA writes the protein side in three-letter codes, so its rows keep counts and places and no letters.
  • A file can hold more than one search of one kind. A file with a protein search and a DNA search, or with parts of two searches pasted together, is refused with the line where it changes. Import each search as its own file.
  • The query keeps the name the file gives it. A query named US-12-744-375-185 counts as SEQ ID NO 185 when the case is on application 12/744,375, so its references go under the feature that recites it.
  • In References, the import's matches name it US patent office search. In Compare searches, its method reads Imported results (US patent office search).
  • A row that does not join stays in the results, with the reason as its title. Rows that did not join lists them with their reasons, and Download them as CSV saves them all.
  • A query named SEQ_ID_NO_<n>, as the search request's FASTA names each SEQ ID NO, counts as that SEQ ID NO: its references go under the claim feature that recites it in References. Other queries' references go under Other searches.
What the case records about an import
  • The case log records the import with the file's name, size, format and SHA-256. The report names the file and its SHA-256, how many rows joined, and the first rows that did not, each with its reason.
  • An import counts as one of the 100 searches you can add to a case, and Remove takes it out again.

The decisions file

Under the report buttons, Download decisions (CSV) and Download decisions (JSON) save every decision in the case:

The file holdsWith
Each claims importThe rules version that read the claims and the set it became, with its label and date, and each change of the current set. Each row also names the set that is current now
Each claim feature as it standsWhere it came from (the rules, a model or you), its rule, kind, claim, quote, details, links and status. The details of allowed changes name the kinds that count, such as "Up to 3 substitutions only"
Each accept, reject and edit, and each feature you addedWho did it, when, and any note. An edit gives the quote, details and links before and after it
Each call to a claims modelThe model, the prompt's ID and SHA-256, each suggestion kept, with what became of its details and its paragraph of the description, and each suggestion dropped, with the reason. A failed call gives its error
Each decision on a drafted sentenceThe last one on each sentence marked
Each decision on a summaryThe template that filled it in, the SHA-256 of the template's words and of the summary's facts, and the last one on each summary marked
Each answer in a claim chartThe claim, feature and reference, your answer and note, and what the rule said then, with the last one on each cell marked
The plan's locked versionsEach draft, edit, lock and unlock, with the reason given for an unlock

Each decision gives the number and SHA-256 of its event in the case log, so you can check it against the log. The JSON file also gives the result of the log's check. Saving the files changes nothing in the case or its log. Only people who can open the case can download them. To try it, download the decisions of the demo case.

The log

Log lists every action in the case, and Verify the log checks how its events tie together:

  1. First event: the first action in the case
  2. Next event: holds the SHA-256 of the event before it
  3. Every later event: the same, back to the first

A changed or deleted event no longer matches the SHA-256 that the next event holds, so Verify the log shows it.

If the server stops while it records an action, Verify the log can say "The case record does not name the last event." The next action you take completes the record, and the log verifies again.

A decisions file you save gives each decision's event number and SHA-256, so you can check the log against it later: see The decisions file.

Hand a case to a colleague

On the hosted service and on a server with sign-in, a case's owner can hand the case to a colleague.

  1. Hand over: in the case header
  2. Continue: after you give your colleague's account
  3. Hand over the case: once the page says what moves
  • Your colleague must have signed in here once, and the server must let them sign in now. On the hosted service, and on a server that names accounts by email address, give their email address, such as b@example.org. It must be at your own email domain. On another server, give the name your colleague signs in with.
  • The case goes with everything in it: its claims and sets of claims, plan and locked versions, runs and their searches, the searches you added or imported, references, decisions, summaries, notes and log. Nothing is copied, and a case has one owner.
  • The case leaves your list, and you can no longer open it, its searches or its listing. Your colleague can hand it back.
More about a handover, and what the record shows
  • Your colleague opens every search and result the case holds, and its listing. Your other searches and listings stay yours. A search of the case that you ran in a combined search or a batch shows there as missing while your colleague holds the case.
  • A case waits while the claims model reads it or drafts for it, or while its listing imports. Hand it over once that is done.
  • The case log records each handover, and the case header and the case's card in your list name the last one, such as "Handed over by a@example.org to b@example.org on 13 Nov 2026." The server's audit log records it too.
  • Handing a case over starts no searches, so it still works after a license or trial ends.
  • A program can hand a case over with an API token: POST /api/cases/:id/handover with {"to": "b@example.org"}.

Listings and the ST.26 check

The Listings tab of Sequence tools reads sequence listings and checks them before they are filed.

  • Import a WIPO ST.26 XML or ST.25 listing, plain or gzip, of up to 256 MiB unless your server sets another size. A gzip file can unpack to at most four times that size. The page compresses a large file before it uploads it. Then pick up to 20 of its sequences, 100,000 letters in all, and choose Start the searches.
  • Import a publication's full-text XML when its listing is inside it, as in the patent publications of 2001 to 2004. Sequences reads the listing inside. If an entry cannot be read, it is left out and the listing names it, such as "SEQ ID NO: 3 could not be read". A publication whose listing was filed separately says so: import that listing instead.
  • Check a listing before you file it: choose a file to check against WIPO ST.26, version 1.7. The check runs in your browser, so the file does not leave your computer. Every problem names its SEQ ID NO and line. An error is a fault that fails ST.26 validation. A warning departs from the standard's wording or advice but passes. Show every problem, errors only or warnings only, filter by SEQ ID NO, and choose Download problems (CSV).

To try them, choose Use demo data beside each file button. To import, it chooses us12788009b2-listing.xml, the ST.26 listing of US12788009B2 with its 16 peptides. To check, it chooses us12421493b2-st26.xml, an ST.26 file as its applicant filed it, cut to 3 sequences, which passes. us20260345831a1-with-listing.xml is a publication's full-text XML with its listing inside.

Antibody CDR search

Turn on Antibody CDRs, beside the search card's title, to find the patents and stored sequences that hold an antibody's CDRs.

  1. Under Search for, choose CDRs of antibody chains and paste up to 10 heavy or light chains of up to 2,000 letters each, then choose Number chains. Human and mouse heavy, kappa and lambda chains can be numbered. Or choose CDRs typed in and type H1 to L3, 1 to 80 letters each. Edit these CDRs changes what the numbering found.
  2. Choose the numbering (Kabat, Chothia or IMGT), a protein set and a Minimum identity from 50 to 100 percent, 80 by default, then choose Search CDRs.
  3. Each CDR runs the exact scan and, where FASTA36 is installed, Smith-Waterman, with up to 5,000 sequences for each CDR.

CDR matches groups the results by patent or by stored sequence, and Only those with every CDR keeps the rows that hold all of them. Each cell reads exact, the identity of the best alignment, or missing. A CDR counts as found at the minimum identity or more over its full length.

On a server with antibody names (the installation guide's Antibody names), each match also names the therapeutic antibodies it shares sequence with: those whose whole heavy or light variable domain the stored sequence holds, or whose three CDRs of one chain it holds letter for letter, in order, under the numbering you chose. By patent, a match names an antibody from all the SEQ ID NOs it lists. Each name links to its page at ChEMBL, and the CSV's last column lists every name with its ChEMBL ID. The names come from ChEMBL, of EMBL's European Bioinformatics Institute (EMBL-EBI), under Creative Commons Attribution-ShareAlike 3.0 (CC BY-SA 3.0). Sources and terms names the release and the credit ChEMBL asks for.

To try it, choose Use demo data beside Open FASTA. It fills trastuzumab's two variable domains from PDB 1N8Z (trastuzumab-chains.fasta) and numbers them, or with CDRs typed in, its six Kabat CDRs.

Sequence tools

Besides Listings, Sequence tools has four tabs:

  • Editor: check, slice, reverse complement and translate a sequence, show its six frames, and Fetch a sequence from the snapshot by accession, or by publication and SEQ ID NO.
  • Fragments: the sequences you saved, each with its edit history.
  • Align: align two sequences by Smith-Waterman or global alignment, or many sequences, up to 200, with MAFFT and a majority consensus.
  • Query tools: queries built from a claim's words. See Query tools.

In Editor, Use demo data fills the coding sequence of human insulin, NCBI RefSeq NM_000207.3 (insulin-cds.fasta). Under Fetch a sequence from the snapshot, it asks for SEQ ID NO: 1 of US12788009B2, granted 2026-11-10, which a snapshot holds only with its additions from that week on. Align takes its sequences from the demo listing or the editor.

Query tools

Each tool says what it read and why it suggests what it does. Nothing is searched until you choose Use in the search form, or search a saved group. Save as a query group keeps a tool's queries together: Query groups, under the tools, lists yours. Search, on a group, chooses the set and method and runs the group's queries as one batch, up to 20 queries in each search. Each search's record names the group and each query.

  • Claim phrase: paste words from a claim, such as "at least 90% identical to SEQ ID NO: 1". The tool reads them with the claim rules a case uses and suggests a search for each sequence: its queries, method, settings and results filter, each with its reason. "At least 90% identical" suggests Global on the query, which counts identity over the whole query, the way such a claim is usually read, with the results opening at Minimum identity 90%. "At least 15 consecutive amino acids" makes each window of 15 a query for Exact text. "With up to 3 substitutions" suggests Within k changes where the server offers it. "X at position 26 is K or R" suggests a Pattern search with [KR] at position 26, and the sequences the choices make. Choose the Listing that holds the SEQ ID NOs the phrase names, or give the Base sequence. Number of the first letter counts the positions from the base's first letter, such as 7 for a sequence that starts at residue 7.
  • Variants: write the changes one variant to a line, as a claim writes them: K26R, Lys26Arg, K26R/Q (one variant with R and one with Q), A8Aib, des-B30, des(1-3), 20insR, K29_T30insR, or 20-30A (one variant for each position from 20 to 30). Save as fragments saves each variant with its changes as a step of its edit history.
  • Windows: cut a sequence into windows of one length and step, into peptides of a range of lengths such as 8 to 15, around positions, or into stretches you name. Where antibody numbering runs, CDRs of an antibody chain cuts out each CDR.
  • DNA pattern: a peptide of 4 to 200 amino acids as DNA, with an IUPAC code for each codon, such as AAR for K, in the standard genetic code or another one. The tool says how many DNA sequences fit the pattern and how many of them code for the peptide. Where one code also allows another residue's codons, as YTN for L also allows TTC and TTT, which code for F, the tool says so, and gives a regular expression that allows only the coding DNA when the pattern search can take it. Search either one with Pattern on a DNA set.
  • Modified residues: paste a peptide in HELM, with its modified residues in square brackets, as H[Aib]EG, or as residues joined by hyphens, or paste the sequence with the listing's notes on its modified residues. Search each modified residue as its parent, such as A for Aib, as X, or both, which makes two queries. Each query's name says which, so the search record keeps the choice.

Hit by several queries, on a group you have searched, counts how many of the group's queries hit each publication in the group's last search, or each stored sequence with Sequences. It lists those hit by at least the number of queries chosen under Hit by, 2 to start, most first, with the names of the queries that hit each one. Download CSV saves the whole list. A query counts once, even when more than one search ran it. Minimum identity (%) and Published before apply to every search as they do to its results, and a patent or alignment you exclude in a search's results does not count. A search that has not finished is left out; choose Count again once it has.

Keyboard and screen readers

Every control on the page can be reached and used from the keyboard.

  • Tab and Shift+Tab move through the controls in page order, and the control with the focus shows a ring. Enter presses a button or opens a link, and Space presses a button or ticks a box.
  • In a row of tabs, such as a result's tabs, a case's steps or Sequence tools, Left and Right move between the tabs, and Home and End go to the first and the last.
  • In a row of chips that takes one choice, such as Query type or Collection, the arrow keys move the choice. Space turns a set's chip under Search in on or off.
  • The keys of the results list are under Results, those of the charts under Charts, and those of the References step under The References step.
  • Esc closes a form that opens in place in a case, as Cancel does, and the focus goes back to the button that opened it.
  • The message before your sign-in ends puts the focus on Stay signed in. Tab moves between its two buttons, and Esc closes it, as Not now does. See When your sign-in ends.
  • Deleting a listing, a saved sequence or a watch, revoking an API token, removing a search from a plan and withdrawing a note each ask first, since none of them can be undone.
  • The status bar and the page's messages are marked as status lines, so a screen reader can read each change without the focus moving. A message stays while the pointer rests on it or while its close button has the focus, and an error stays until you close it.
  • Zoomed to 400%, or on a screen 320 px wide, the page needs no sideways scrolling. Wide tables and alignments scroll inside their own boxes.

My work

On the hosted service and on a server with sign-in, My work lists your searches, the result files you imported into cases, your cases and your watches on one page, newest first. Open it with My work at the top of the search page, or of the docs and the site's other public pages once you are signed in.

  • Show keeps one kind: searches, imports, cases or watches. Each choice says how many there are.
  • Each card says what its card on the search page says, such as a search's state and hits, a case's stage or a watch's last run. Open goes back to it: a search or an imported result file to its result page, a case to the case, and a watch to Watches on the search page.
  • A search of several sets at once shows as one search. Exact alignments, multiple alignments, listing imports and portfolio exports are not listed there; they stay under Recent jobs on the search page.
  • On a free trial, the page shows the days left, as the search page does.
  • The page lists only your own work and changes nothing. The local workbench has no accounts, so it has no My work page.

Your account

Once you are signed in, your account button sits at the top right of every page: a circle with your initials, and your first name on a wider screen. Its list has My work, Account, API tokens if your server allows them, Docs and Sign out. An admin also sees Usage, which counts every account's work on the server.

Account opens your account page, with these tabs:

  • Profile: your name and email address. If you sign in with a ConductScience Account, Manage your ConductScience Account opens that account's page, with its details and sign-in settings. If you sign in through your organization, Profile shows your user name instead, and you change your details through your organization's sign-in.
  • Plan: your free trial's days left and the plans to choose from, or your organization's license. With neither, there is no Plan tab.
  • API tokens: the tokens your programs sign in with. See Programs and API tokens.
  • Email alerts: your watches, each with its state and where its emails go. Click a watch's name to pause, resume or delete it on the search page.
  • Usage: the searches you started each day over the last 7, 30 or 90 days, by UTC date. Result files you import are not counted.

The local workbench has no accounts, so it has no account button or account pages.

Who can see your work

  • On the hosted service and on a server with sign-in, your searches, portfolio exports, cases, listings, saved sequences, watches and applicant joins belong to your account. No other account can open them, except a case you hand to a colleague, with the searches and listing it holds. See Hand a case to a colleague.
  • An admin's usage page counts each account's searches and other work. It shows counts, not content.
  • An admin can also download a file of the decisions in every case, to measure how well the claim rules and model prompts work: each feature's kind, rule, source and status, each model call's counts and the reasons suggestions were dropped, in fixed words. The file holds no claim words, quotes, notes or edits, since claims pasted into a case can be from an application that is not yet published.
  • The people who run a server hold its stored files: ConductScience for the hosted service, and your organization's staff for a server it runs.
  • You can delete a listing, a saved sequence, a watch or a demo case you started. Searches, portfolio exports and other cases have no delete of your own, on the page or through the API. An admin can delete a case with its searches: the case and all it held are gone, and the server's audit log keeps the case's ID, its owner, who deleted it and when, and a fingerprint of its log, never what it held.
  • Signing out deletes what Sequences was still making for that browser: a listing or result file still uploading and an export still being sent. The page also clears what it kept in the browser, such as a closed Things to try. Your searches, listings, cases and other work stay. A sign-in that ends on its own does the same within a minute, and signing out in one browser leaves your other browsers as they are.
  • A search keeps its results and record. The files it made on the way, such as its query in parts, are deleted once it finishes, fails or is cancelled.
  • A server's admin can set how long records are kept. Searches that no case holds, portfolio exports and batches are then removed that many days after they finish. A search a case holds, or a watch follows, stays. Unless the admin sets this, records stay.
  • On the local workbench there are no accounts: everyone who can open the page sees all the work saved there.
  • Claim text and a case's facts leave the server only when you ask for model suggestions or a draft, and then they go to the model the server names: for suggestions, the claims with the paragraphs of the description chosen for them; for a draft, that text's facts. The case keeps a record of each call: for suggestions, what was sent, the model's answer and each suggestion kept or dropped with the reason; for a draft, the model, the prompt and what became of the draft: accepted, edited, rejected or dropped by a check.
  • Where a server sends watch emails, each goes through Resend, an email delivery service. It holds your address, how many new matches or publications one of your watches found and its links back to the server, and names no watch, query or match, unless the server's staff choose emails with details, which also carry the watch's name and its new matches or publications.
  • The ST.26 check runs in your browser and sends nothing.

Programs and API tokens

Programs can run searches through the same routes the page uses.

  • On the hosted service, and on a server whose sign-in takes API tokens, choose API tokens from your account button. Name the token and pick how long it lasts, up to 90 days (your server may set less). Copy it right away: the page shows it only once. An account may have 10 live tokens. The page shows when each token was last used, and Revoke ends one. Use demo data fills the name Demo program and the shortest time.
  • The program sends the token with every request, as Authorization: Bearer <token>, and acts as your account under the same rules.
  • GET /api/openapi.json describes the routes a program uses, in OpenAPI 3.1. It is open without sign-in.
  • POST /api/batches starts 1 to 100 searches in one request. If any of them cannot run, none starts, and the answer lists each problem. GET /api/batches/:id gives the state of each search. An account may have 200 batch searches waiting or running. api-batch.json is a demo body of two exact text searches: add a snapshotId from GET /api/databases to each search before you send it.
  • GET /api/batches/:id/multi-hits counts a batch's finished searches together, as Hit by several queries does for a query group: by=publication or by=sequence, and minQueries, how many queries must hit a row for it to be listed, 2 unless set. identity, coverage, queryMatch, queryIdentity, hitIdentity, publishedFrom, publishedBefore and office apply to each search, and GET /api/batches/:id/multi-hits.csv gives every row. A query group's search is one batch, named by batchId in the group's searches.
  • DELETE /api/cases/:id deletes a demo case you started. A case that is not a demo case answers 409 and stays.
  • On a search's patents, GET /api/jobs/:id/publications takes group=family for one row per patent family, at its earliest publication, with its members and how they were joined, and later=hide to fold the publications from on or after the Published before date and count them. GET /api/jobs/:id/applicants and GET /api/jobs/:id/applicants.csv take later=hide too, and then count only the publications that stay. POST /api/jobs/:id/marks takes ids to keep or exclude up to 500 publications at once, such as a family's. The OpenAPI description gives every field.
  • GET /api/jobs takes text to find your searches by name, or by the set or database they searched, and status to keep those in the states you list, such as queued,running. GET /api/jobs/:id/request gives the request a search was started with, query included.
  • POST /api/groups runs one search across 2 to 8 sets as a combined search, such as a protein query against Patent proteins, Patent DNA and Swiss-Prot. If one set cannot take the query, or the queue lacks room, none starts. GET /api/groups/:id/references gives one list: a patent or database entry that several sets found is one row, with its best measures, its earliest date and the sets that found it. It takes the filters, sorts, group=family and later of a search's patents, plus set and kind. A mark through POST /api/groups/:id/marks reaches every search in the combined search. On the page, Search in starts one when you check two or more sets.
  • GET /api/jobs/:id/hits/:row gives one hit of a search by its row number, whatever the filters, with its mark.
  • On several servers that share one queue, a search of a snapshot has shared: true. GET /api/jobs/:id adds ahead while it waits: how many such searches wait before it. worker names the server that runs it or ran it, and retries counts the times it was cut off. GET /api/status adds servers: the other servers that take searches now, and how many searches of a snapshot run at once.
  • POST /api/cases/:id/imports?name=<file name> brings a result file into a case, as Import a result file does. Send the file itself as the body, as application/octet-stream, or as application/gzip when it is compressed. GET /api/cases/:id/imports/:jobId/unresolved lists the rows that did not join, with their reasons. For a file whose columns you match, add columns= with a JSON column map, URL-encoded, such as {"hit":"Subject","identity":"% identity"}, and if you like header= with the header row's line and delimiter= with comma or tab. GET /api/cases/:id/imports/:jobId/report gives such an import's column map and its import report.
  • A token cannot make another token: that needs the browser. On the local workbench there are no tokens, and a program on the same computer calls http://127.0.0.1:8765 as the page does.
  • To send an application's documents into cases from another system, use a connector token: see Connectors.
  • An assistant on your computer can call these routes for you through the assistant program: see Assistants.

Assistants

An assistant that supports the Model Context Protocol, an open standard for connecting assistants to tools, can run searches and read their results through the assistant program. The program runs on your computer and calls this server with your API token, so each call acts as your account under the same rules as the page. The program runs no model: your assistant reads the answers on its side. What the assistant reads goes wherever that assistant sends it, so follow your organization's rules for the assistant you use.

Download sequences-mcp.mjs and run it with Node.js 22 or later. It needs nothing else. Make an API token as Programs and API tokens describes, save it in a file by itself, and check the address and the token from a terminal:

node sequences-mcp.mjs --server https://sequences.example.org --token-file sequences-token.txt --check

Then add the program to your assistant's tools, as the assistant's own instructions say. Most take a command and its arguments, such as:

{"command": "node",
 "args": ["<folder>/sequences-mcp.mjs", "--server", "https://sequences.example.org", "--token-file", "<folder>/sequences-token.txt"]}

On the local workbench, leave out --token-file: a program on the same computer needs no token. The assistant then has seventeen tools:

  • list_sets and list_snapshots: the sets this server searches, and each snapshot's sources and terms.
  • start_search and start_batch: one search, or 1 to 100 at once, with the fields of POST /api/jobs.
  • get_job and get_batch: the state of a search, or of each search in a batch. Each can wait up to 45 seconds for the search to end.
  • get_hits: a finished search's hits, a page at a time, with the filters and sorts of the results page.
  • get_publications: the patents a search found, or every publication of one hit's sequence.
  • get_multi_hits: the patents or stored sequences that several of a batch's queries hit, as Hit by several queries counts them for a query group.
  • get_patent_text: What the patent says about the SEQ ID NOs that a search matched in a US publication.
  • get_export: a search's CSV, FASTA and text exports, as text. A file longer than 50,000 characters is cut at the end of a line, unless the assistant asks for more, up to 1,000,000.
  • start_cdr_search and get_cdr_matches: an antibody CDR search and its CDR matches.
  • save_query_group, list_query_groups, get_query_group and start_group_search: query groups, and a group's queries searched as one batch.

A search the server would ask about first, such as a long one, answers with its estimate and does not start. The program tells the assistant to ask you, and to send it again with confirmed: true only if you agree. The program never prints the token.

Connectors

A connector lets another system, such as your document system, send an application's claims and sequence listing into a case, so the case is ready when the searcher opens it. Each call sends one document set: a manifest, which says what the set is and which account owns the case, and the files it lists. The files go in through the same importers as on the page, so the case opens with its claims, the features the rules found and its listing.

Make a connector token

An admin of the server makes connector tokens. On the hosted service, write to support@conductscience.com for one.

  1. Choose API tokens from your account button. Below your own tokens, Connector tokens lists the server's.
  2. Name the token after the system that will use it, pick how long it lasts, and list the accounts it may make cases for, separated by commas or new lines. Each account must have signed in to this server once.
  3. Choose Make connector token and copy the token into a file the system can read, such as connector-token.txt. The page shows it only once.
  • The token can call only POST /api/connector/imports. It cannot search, open a case or make another token.
  • It makes cases only for the accounts you named, and only while you are an admin here. A case belongs to its owner, so you do not see it unless it is yours.
  • A server may have 20 live connector tokens, each for up to 50 accounts. The list shows when each was last used, and Revoke ends one.

The manifest

The manifest is a JSON object with these fields:

  • source: the sending system's name, up to 40 characters: letters, digits, spaces, dots, dashes and underscores.
  • sourceId: that system's own ID for the document set, such as a docket number, up to 100 characters. A set with the same source and sourceId always goes into the same case.
  • title: the case's name, up to 200 characters.
  • owner: the account that will own the case, as it signs in. The token must be allowed to name it.
  • applicationNumber and priorArtDate (YYYY-MM-DD), if you want to set them, with priorArtDateBasis: filing, priority or other, the default. Without them, a publication's XML gives the case its application number, and its filing date as the prior art date.
  • files: 1 to 5 files, each with its name (the file name alone, without a folder), its role and its sha256, the file's SHA-256 as 64 hexadecimal digits.
  • version: 1, if you give it.

Each file has one of three roles:

  • claims: a file of claims, as XML, a Word file (.docx) or text, up to 15 MiB.
  • listing: a WIPO ST.26 XML or ST.25 listing, plain or gzip, as Import on the Listings tab takes. A case has one listing.
  • publication-xml: one publication's full-text XML from the US patent office, an application or a grant, up to 15 MiB. Its claims become a set of claims in the case. When the manifest lists no listing and the XML holds one, that listing becomes the case's.

A complete manifest for a grant's XML and its listing:

{
  "version": 1,
  "source": "Docketing",
  "sourceId": "PAT-118",
  "title": "Insect neuropeptides, PAT-118",
  "owner": "ana@example.org",
  "files": [
    { "name": "PAT-118.xml", "role": "publication-xml", "sha256": "<SHA-256 of PAT-118.xml>" },
    { "name": "PAT-118-listing.xml", "role": "listing", "sha256": "<SHA-256 of PAT-118-listing.xml>" }
  ]
}

Work out each SHA-256 with sha256sum PAT-118.xml on Linux, shasum -a 256 PAT-118.xml on a Mac or Get-FileHash PAT-118.xml in PowerShell, which prints capitals: the manifest takes either.

Send a set

Send the manifest first, as the part named manifest, then each file as a part named file with its name from the manifest, all in one multipart/form-data request. With curl, from the folder that holds the files, and the token in SEQUENCES_CONNECTOR_TOKEN:

curl -H "Authorization: Bearer $SEQUENCES_CONNECTOR_TOKEN" \
  -F "manifest=<manifest.json;type=application/json" \
  -F "file=@PAT-118.xml" \
  -F "file=@PAT-118-listing.xml" \
  https://sequences.example.org/api/connector/imports

To try it, save connector-manifest.json as manifest.json in one folder with us12788009b2.xml and us12788009b2-listing.xml, and give the account the case is for as owner. The case it makes is an ordinary case, not a demo case, so only an admin can delete it.

The answer

The answer is 201 when the call made the case, and 200 when the set went into the case its source and sourceId made before:

{
  "case": {
    "id": "<case ID>",
    "title": "Insect neuropeptides, PAT-118",
    "owner": "ana@example.org",
    "made": true,
    "url": "https://sequences.example.org/?case=<case ID>"
  },
  "source": "Docketing",
  "sourceId": "PAT-118",
  "files": [
    { "name": "PAT-118.xml", "role": "publication-xml", "sha256": "<SHA-256 of PAT-118.xml>",
      "result": "imported", "claims": 9, "features": 3, "set": 1 },
    { "name": "PAT-118-listing.xml", "role": "listing", "sha256": "<SHA-256 of PAT-118-listing.xml>",
      "result": "imported", "sequences": 16, "listingId": "<listing ID>" }
  ]
}

Each file's result is one of:

  • imported: it went in. A file of claims gives its claims, the features the rules found and the set it became. A listing gives its sequences and listingId.
  • already: the case holds it from an earlier call. The same set sent again changes nothing, so a system can send a set again whenever it is unsure the first call arrived.
  • refused, with its reason: it did not go in. The other files did.

Each step in the case log names the token, such as "Docketing (connector)". The server's audit log records every call with its source, its ID and each file's SHA-256 and result.

What a refusal looks like

When the whole call is refused, nothing goes in, and the answer gives the reason as message:

403 Forbidden
{
  "message": "This connector token may not make cases for ben@example.org. An admin names the accounts it may make cases for when they make the token.",
  "source": "Docketing",
  "sourceId": "PAT-118"
}
  • 400: the manifest cannot be read, or a file is missing, not in the manifest or not the file its SHA-256 names. The message says what to change.
  • 401: the token is wrong, expired or revoked.
  • 403: the token may not name the owner, or the admin who made it is no longer an admin.
  • 409: the case these documents went into before belongs to another account now.
  • 413: a file, or the whole request, is larger than the server takes.
  • 415: the request is not multipart/form-data.
  • 429: the search queue is full, so the listing cannot import now. Send the set again in a few minutes.

When one file is refused, the others go in, and its line gives the reason:

{
  "name": "PAT-118-listing-2.xml",
  "role": "listing",
  "sha256": "<SHA-256 of PAT-118-listing-2.xml>",
  "result": "refused",
  "reason": "The case has a listing already. Choose another in the case if this one should replace it."
}

When no file can go in, no case is made: the answer is 400, "No file could be imported, so no case was made.", with a line for each file. Fix what the message names and send the set again.

The folder program

Download sequences-connector.mjs, the reference program, and run it with Node.js 22 or later on a computer where your documents arrive and that can reach the server. It needs nothing else.

node sequences-connector.mjs watch --inbox /srv/sequences-inbox \
  --server https://sequences.example.org --token-file /etc/sequences/connector-token.txt
  • Each folder in the inbox is one set: its manifest.json and the files it lists. The program sends a set once every file it lists is there with its SHA-256, so a set still being copied in waits. A set that is still not whole after 10 minutes (--wait, in seconds) goes to failed/ unsent.
  • After the answer, the folder moves to done/ when every file went in or was in already, or to failed/ when the server refused the set or a file, with the answer beside the files as result.json. To send a failed set again, fix it and move it back into the inbox.
  • With no answer, or a 429 or 5xx answer, the set stays and goes again on the next round. A token the server no longer takes stops the program.
  • It reads the inbox every 30 seconds (--every), or once with --once. It never prints the token.

It prints a line for each set:

Sequences connector adapter 1.0.1: watching /srv/sequences-inbox, sending to https://sequences.example.org, every 30 seconds.
done PAT-118: 201, case <case ID> (new), 2 imported, 0 already in, 0 refused. Moved to /srv/sequences-inbox/done/PAT-118.

Data sources and dates

Each set is built from public files, and Sources and terms on the search page names the release, the terms and the notices of each one. The page lists each set from the newest snapshot that holds it. On the hosted service, the patent sets come from the snapshot of November 13, 2026, built from GenBank release 274.0, the patent protein files of EMBL-EBI, and the US patent office's weekly full-text files and sequence listings of US published applications and patents since 2001. Swiss-Prot, TrEMBL and the ESTs come from the snapshot of November 10, 2026; the ESTs are from ENA's files of September 10, 2026. NCBI core_nt is outside the snapshots, and a search of it names the date of its copy under the search's name. Paper citations, dates and marks come from PubMed records revised up to November 8, 2026 and the Retraction Watch Database of November 6, 2026, and the genes and diseases a paper names from PubTator 3 of October 6, 2026. Applicant suggestions come from ROR v2.15, data of October 20, 2026. Where genes run, they come from NCBI Gene and HGNC of November 11, 2026, through UniProtKB/Swiss-Prot of November 4, 2026, in ConductGraph's Gene search 0.2.1. A server your organization runs has the data its staff loaded, and Sources and terms names it.

Limits

WhatLimit
Queries in one search20, of up to 20,000 letters each and 100,000 in all
BLAST and FASTA36 results5,000 sequences for each query on a set of the snapshot; BLAST keeps 5 alignments for each sequence
Exact and pattern results1,000,000 rows in the list; the totals count every match
Within k changesQueries of up to 64 letters; k from 0 to 5, below the query's length; 50,000 hits listed for each query, and the counts are complete
Size of one search's results8 GiB, unless the server sets another size; 256 MiB for each run of a FASTA36 program. A search that passes its limit fails, and the message says what to change. An exact or pattern list can also end before 1,000,000 rows because of the limit, and the totals still count every match
Align exactly1,000 rows at a time
More downloads1,000 selected rows at a time; a FASTA file stops before 100,000,000 letters
Aligned letters in a workbookThe first 10,000 rows of Alignments; up to 32,767 letters in a cell
A pattern100 parts, 4 ranges, repeat counts up to 1,000 and 20,000 characters; a regular expression up to 8 choices, each within these
An antibody search10 chains of up to 2,000 letters, or typed CDRs of 1 to 80 letters each
A sequence listing256 MiB unless the server sets another size
Time for an upload to arrive30 minutes, for a sequence listing, a result file imported into a case or a connector's documents
A search of a listing20 sequences and 100,000 letters
A collection you import20 MiB and 100,000 sequences, on the local workbench, with up to 500 hits for each query
Watches50 for each account, of searches and applicants together, with up to 50,000 matched records in a watched search
A portfolio export100,000 sequences and 100 million letters
API tokens10 live tokens for each account, each for up to 90 days
Batches100 searches in a batch, and 200 waiting or running for each account
Hit by several queries200,000 publications or sequences in one count
A run of a case's plan100 search jobs; a larger plan runs as several runs, one after another
A result file imported into a case200 MB and 1,000,000 rows
Case reports100 searches added or imported; up to 500 notes in a case, each up to 5,000 characters; and 100 applicants under By applicant, with up to 20 publications each
Sentences you edit2,000 characters each
Applicant joins2 to 50 names in a join, and 5,000 joins for each account
Time for one search or exportOne hour, unless the server sets another limit
Processor timeNone, unless the server sets CPU hours for each account in a UTC day. See Searches that ask first

BLAST can miss matches that an exact scan or Smith-Waterman finds. These results do not establish patentability. Questions about Sequences go to support@conductscience.com.