Sequences

This site runs Sequences 0.116.0

Release notes

What each release of Sequences changed, newest first. The hosted service and the server edition run the same releases. Each release names the version of its search engine, the part that runs searches, and every search record names the engine version that ran it.

Sequences 0.116.0

Search engine 0.21.0

Import a result file now reads the US patent office's sequence search results, so an examiner's searches can sit in a case beside Sequences' own.

  • New: Result files from other programs reads the US patent office's sequence search results, as office actions print them: protein and DNA searches, among them those that translate DNA. Each hit comes in rank order with its Query Match, score and alignment, and joins the patent sequence its record names, by publication and SEQ ID NO or by accession.
  • New: in References, such an import's matches name it US patent office search, and in Compare searches its method reads Imported results (US patent office search). A result file looked up in one patent set now compares with a search of that set, sequence by sequence.
  • New: in these results, a query named US-12-744-375-185 counts as SEQ ID NO 185 in a case on application 12/744,375, so its references go under the feature that recites it.
  • New: POST /api/cases/:id/imports reads US patent office search results. Their rows keep each hit's rank and Query Match, and the import's record keeps each search's header under imported.searches.

Sequences 0.115.0

Search engine 0.21.0

Sequences moved to sequences.conductscience.com. The old address, sequences.conductscience.org, forwards there, and API tokens work at both.

  • If you were signed in at the old address, sign in once more.

Sequences 0.114.0

Search engine 0.21.0

A case's Report step now charts each claim's features against its top references, and the charts go into the report and a CSV.

  • New: Claim charts. Each independent claim gets a table of its features against its 5 top references, kept ones first. Fixed rules mark each cell Met, Not met or Not shown, with the identity, aligned positions and SEQ ID NO, each linked to its source. Set any cell to Met, Not met or Unclear, with a note. The case log keeps each answer.
  • New: the case report prints each chart, Download CSV saves every cell, and the decisions file lists your answers.
  • New: GET /api/cases/:id/charts, POST /api/cases/:id/charts, POST /api/cases/:id/charts/cells and GET /api/cases/:id/charts.csv.

Sequences 0.113.1

Search engine 0.21.0

0.113.0 with three fixes, two of them found by testing in Firefox. On the hosted service, the Accessibility page now reports a test with the NVDA screen reader, and says the accessibility tests run in Chromium and Firefox.

  • Changed: on the hosted service, the Accessibility page now says that on release 0.112.0, NVDA, a screen reader for Windows, read the search page and a case in Firefox while a script ran a search, read its results and took the case from Claims to Report, and that what NVDA said at each of the 38 steps was recorded.
  • Changed: each release's accessibility tests now run in Firefox as well as Chromium, on Windows 11, and the Accessibility page says so.
  • Changed: the Accessibility Conformance Report, now version 1.6.0 for release 0.113.1, rates 302.1 Without Vision as Supports after that NVDA pass.
  • Fixed: under Sequence tools, a query group's Search could open nothing if pressed while the list of groups was updating.
  • Fixed: in Firefox, with wider text spacing, the Best rank column under All references could cut off its text.
  • Fixed: on the sample case's pages, a patent too new for Google Patents now opens at Espacenet, or as the European Patent Office's PDF for a European one.

Sequences 0.113.0

Search engine 0.21.0

The benchmarks page opens with its answers: six questions, each with Yes or Mostly and one figure, and a table of the search that fits each task.

  • Each test is headed by the question it answers, with the release and date it ran on and what it means for a patent search, and keeps its full numbers in a fold. A list of terms closes the page.

Sequences 0.112.0

Search engine 0.21.0

Applicants now follows your dates as Patents does: with Published before set, it folds the later publications and its counts leave them out.

  • Fixed: with Published before set, Applicants now folds the later publications under its list, as Patents does, and its counts leave them out until you choose Show them. Its CSV and the Applicants sheet of the Excel workbook hold what the tab shows.
  • New: GET /api/jobs/:id/applicants and applicants.csv take later=hide, as GET /api/jobs/:id/publications does. Without it, they count every publication, as before.

Sequences 0.111.4

Search engine 0.21.0

The same service as 0.111.3, with search times measured again on November 20.

Sequences 0.111.3

Search engine 0.21.0

The user guide now gives the size limit of one search's results and the time an upload has to arrive, and the installation guide corrects the memory the Modifications add-on takes.

  • New: Exact and pattern search and Limits in the user guide give the size limit of one search's results: 8 GiB, unless the server sets another size; 256 MiB for each run of a FASTA36 program. A search past its limit fails, and the message says what to change. An exact or pattern list can also end early because of it, and the totals still count every match.
  • Fixed: the installation guide's Modifications section said the folder's table takes two to three times the size of records.tsv in memory. It takes up to twice that size, and up to three times while the service reads the folder at start.
  • New: Limits in the user guide give the time an upload has to arrive: 30 minutes, for a sequence listing, a result file imported into a case or a connector's documents.

Sequences 0.111.2

Search engine 0.21.0

Searches with thousands of hits finish much sooner, and a cancelled search no longer holds up the searches waiting behind it. The hosted service now shows modified residues and bases on hits and patents, from 1,547,849 patent sequence records. The hosted service goes from 0.110.0 to 0.111.2, so 0.111.0's changes reach it too.

  • Searches that keep thousands of hits finish much sooner, with the same results. On 20 November, a BLAST search of Patent DNA for a query of 30 bases that kept 5,000 hits took 98 seconds on the hosted service, most of it spent finding which claims name each hit's SEQ ID NO. In a test with the hosted service's own data, that step now takes under a second instead of 66.
  • Cancelling a search now stops its work at once. Before, a cancelled search with many hits could keep the server busy for minutes, and searches waiting behind it waited too.
  • New on the hosted service: the Modifications add-on of 0.111.0, which shows modified residues and bases on hits and patents. Its folder holds the 1,547,849 records that mark them in GenBank's patent sequences, EMBL-EBI's US patent proteins and the US patent office's sequence listings (its weekly full text files since 2001 and its PSIPS downloads), read on 17 November 2026.
  • Fixed: 0.111.0 and 0.111.1 could not open a Modifications folder as large as the one read on 17 November 2026, with 1,547,849 records. 0.111.2 opens it.

Sequences 0.111.1

Search engine 0.21.0

0.111.0 with one test fixed so that it installs. The service's code is 0.111.0's.

  • Fixed: one of 0.111.0's tests ran a command that the service's test image does not hold, so 0.111.0's install stopped at its tests. The service's code is 0.111.0's.

Sequences 0.111.0

Search engine 0.21.0

Compare searches can now list the patents that 2 to 6 finished searches found, each patent once. The search page now has its own address, /search. Results now keep the source of each record. Where a server runs the Modifications add-on, hits and patents show the residues and bases their records mark as modified.

  • New: four more use-case pages show how Sequences fits a search: Prior art and novelty; Sequence variants; Law firms and search firms; Technology transfer offices.
  • New: five guides cover sequence listings and how to search patent sequences: ST.25 and ST.26 compared; ST.26 checklist; Patent sequences in BLAST; Reading a sequence listing; Buyer's checklist. The list of all pages has a new Guides group.
  • New: the Data sources page gives each patent set's size in three units: records; publication and SEQ ID pairs; distinct sequences.
  • Compare searches by patent: choose Patents under the searches compared to see one list of the patents 2 to 6 finished searches found, each listed once with what each search found in it. Minimum identity, Published before and Patent office narrow every search alike, and the part shown downloads as CSV or Excel. Patents match by office, number, kind code and publication date, even across sets.
  • The search page now has its own address, /search. Links to the old address still work.
  • Reports, workbooks, case plans, watch emails and the API reference now say additions, not weekly additions, where they name a snapshot's additions. In the workbook's Search record, the row is now Additions.
  • Results now keep the source of each record. When a source's terms keep its sequences on the hosted service, files carry its result rows and credit but not its letters. FASTA, alignments.txt and the workbook say how many rows have no letters, and downloads count them in new headers. BIOSEQUENCE_CONFIDENTIAL_SETS also takes a source's ID, to keep its rows out of external formats.
  • Modified residues and bases on hits and patents. Where a server runs the Modifications add-on, a hit whose patent records mark modified residues or bases shows them in the records' own words, such as Aib at position 2, in its pane and in a Modifications column on Alignments and Patents. Two filters keep the modified hits, or those with a modification you name, and exports get the new columns.

Sequences 0.110.0

Search engine 0.21.0

The search page now starts on Main plus New wherever a set has additions. Watch emails now say only how many new matches or publications one of your watches found, and name no watch, query or match. Server staff can choose emails with details instead.

  • The search page now starts on Main plus New wherever a set has additions, so a search reads every record added since the snapshot was built unless you choose Main. Searches through the API still read Main unless they name the additions.
  • Fixed: the time estimate for a search of Main plus New now counts the added records as well as the snapshot.
  • A watch email says only how many new matches or publications one of your watches found, such as "One of your watches has 3 new matches. Sign in to see them.", with a link to Your watches and a link that pauses the watch and its emails. It names no watch, query or match: what the watch found shows under Your watches after you sign in. See Watches.
  • For server staff: a server that sends watch emails sends this shorter email after the update. BIOSEQUENCE_MAIL_DETAILS=on keeps the email with details, as before: the watch's name, its set or applicant and up to 50 of its new matches or publications. Any other value stops the service at start. See Watch emails.
  • Every docs page now ends with a list of the docs pages, the one you're on marked. On a wide screen the list stays beside the page as you scroll.
  • On the hosted service, the home page offers Start free trial, and the sign-in page tells a new visitor to choose Create an account on the next page.
  • On the hosted service, every page's footer now links to Security and Accessibility.
  • Stored IUPAC codes, an option of Exact text and Pattern: check Let stored IUPAC codes match the letters they stand for, and a code in a stored sequence, such as R for A or G, matches each letter it stands for, up to the most you choose in one match, 3 to start. The Stored codes column and the pane give each match's codes, and the CSV adds stored_codes. With the box clear, searches run as before.

Sequences 0.109.0

Search engine 0.21.0

The hosted service now runs Claimed sequences: the Claimed column says whether claims of a hit's US applications and patents name its SEQ ID NO, for publications through 12 Nov 2026. It also runs Patent families and Patent status. Patent words is a new filter and Variants a new results tab. Up to 2,000 pasted or dropped queries now save as a query group.

  • Patents grouped by family can now use the patent families that Google Patents Public Data gives, on servers that run the Patent families add-on. A family's pane counts its publications and offices from 2001 on and links the ones your result does not hold. A search keeps its families. The Patents CSV and the workbook's Families sheet end with family_id, family_publications and family_offices.
  • Patent status, an add-on: where a server runs it, the US status column gives the group of each US application of a hit, such as Granted or Abandoned, with the US patent office's words and status date in its tooltip, as read from the office's Open Data Portal on the day it shows. The US status filter keeps the hits with at least one US application in a group.
  • The Patents tab, the hit's pane and the CSV and Excel exports carry the same, with the day read. Granted means the office's words read Patented Case, which the office keeps after a patent's term ends, so a granted patent may no longer be in force. Sequences does not estimate when a term ends.
  • Patent words, a new filter on a server with patent context: it keeps the hits and patents whose US publications use every word you type, in the text Sequences keeps from them. The CSV exports and the workbook follow it.
  • Variants: a new results tab draws one query's first 100 hits, with the Alignments list's filters and sort, in the query's own positions. Each position counts the hits that differ and names each change, such as K45R or K45del, and each of those hits opens in Alignments. Export positions (CSV) saves a line per position; programs use GET /api/jobs/{id}/variants. See Variants.
  • More than 20 queries: pasted or dropped queries, up to 2,000, and a listing's picked sequences save as a query group, which runs as one batch that Hit by several queries can count. See Start a search.
  • The assistant program 1.2.0 has nine more tools. An assistant can count what several queries of a batch hit, quote what a US publication says about its matching SEQ ID NOs where the server has patent context, read a search's CSV, FASTA and text exports, run an antibody CDR search and read its matches, and save, list, read and search query groups. See Assistants.
  • To get the new tools, download the program again from sequences-mcp.mjs. A file the assistant reads is cut at the end of a line past 50,000 characters, unless it asks for more, up to 1,000,000.

Sequences 0.108.0

Search engine 0.21.0

Hit identity joins Query identity, each measure now has one name across the results page, the PDF report, the exports and the API, and Query Match, Query identity and Hit identity each set a minimum. Every page now has the same top bar, signed in or not. On the hosted service, the menu now has Use cases, the ST.26 check is under Features, and its page opens with a large file box.

  • New: Hit identity, identical residues as a percent of the whole stored sequence, as Query identity counts them against the whole query. Each measure now has one name on the results page and in the PDF report, one column in the CSV and the workbook, and one field in the API, and Coverage on the results page is now Query coverage. See Results.
  • New: Query Match, Query identity and Hit identity each set a minimum on the results page where the search gives that measure, in saved views, in the workbook and through the API, GET /api/batches/{id}/multi-hits included, and Hit identity sorts Alignments. The guide's Results section gives each measure's formula and which methods give it.
  • The search page, your results, My work and your account pages now have the same top bar as the rest of the site. Once you sign in, the menu shows Search, My work, Examples and Docs, with your account button on the right. Back to search is gone, since Search is always in the menu.
  • On a wide screen the menu fits on one line in the top bar. On a tablet it has its own row under the logo, always in view. On a phone, tap Menu to open it.
  • New on the hosted service: Use cases in the menu lists Freedom to operate and Antibody CDRs in patents. The ST.26 check is now the last page under Features.
  • On the hosted service, the ST.26 check page now opens with a large file box. Click anywhere in it to choose your file, or drop the file on it. Just under the box are the lines that say your file stays on your computer and the page loads no analytics. Feature and use-case pages now end with a panel of next steps. Use-case pages also repeat the panel's first two buttons under the heading.

Sequences 0.107.0

Search engine 0.21.0

Portfolio exports and searches of Main plus New finish sooner and use less memory once a large delta is loaded, because they now share one copy of each delta's index. load and verify no longer fail on a delta whose records file is over about 512 MiB. Our US data load also reads sequence listing files it used to reject, and a later data update adds 155 publications from them.

  • Fixed: each search of Main plus New, each portfolio export and each download of hit sequences loaded its own copy of every delta's index into memory, and the service stopped answering other requests while each copy loaded. For a delta that adds or changes 10.5 million sequences, each copy took about 4 seconds to load and used about 1 GB.
  • The service now keeps one copy of each delta's index, shared by every search, export and download that needs it. For the same delta, that copy loads in about a tenth of a second and takes about 168 MB: 16 bytes for each sequence the delta adds or changes and 16 bytes for each publication it lists. See What you need.
  • Fixed: load and verify failed on a delta whose records.tsv was larger than about 512 MiB, because one check read the whole file into memory at once. The check now reads the file in blocks and checks the same records as before. From now on, a delta needs this release or later, so check which release your server runs before you copy one: see Load a copy.
  • Our US data load now reads three kinds of sequence listing file it used to reject: text files from before ST.25, ST.25 files with fewer sequences than declared or a faulty entry (now only that entry is left out), and ST.26 listings in one text layout. Uploads are read as before. A later data update adds 155 publications, from 2002 to 2024, with about 3.5 million SEQ IDs from these files.

Sequences 0.106.0

Search engine 0.21.0

An AI assistant can now run searches through a program on your own computer. A search's Excel workbook gains an Aligned letters sheet, a searched query group lists what several of its queries hit, and a DNA query with IUPAC codes can be searched as a pattern. Where a server runs the Claimed sequences add-on, each hit says whether claims of its US publications name its SEQ ID NO.

  • The assistant program: an assistant that supports the Model Context Protocol can list the sets, start searches and batches of up to 100, follow them, and read a finished search's hits and patents a page at a time, with your API token. Each server hands it out at /docs/sequences-mcp.mjs. See Assistants.
  • The program runs on your computer and calls the server as your account. It runs no model, and a search the server would ask about first does not start until the assistant sends it again, which the program tells it to do only if you agree.
  • Aligned letters in the workbook: for a search that aligns, Export all tabs (Excel) and a sample's Download as Excel add an Aligned letters sheet after Alignments, with the first 10,000 rows of Alignments in order. Each row gives the letters of the query, the match line and the hit as its pane draws them, and the exact alignment's where Align exactly made one. See Exports.
  • A longer alignment than a cell holds is cut at 32,767 letters in all three columns, and the Search record sheet counts the cut cells.
  • Hit by several queries: a query group you have searched lists the publications, or the stored sequences, that at least 2 of its queries hit in its last search, or as many as you choose, most first, with the names of those queries. A query counts once, even when more than one search ran it, and Download CSV saves the whole list.
  • Programs get the same count for any batch of searches from GET /api/batches/{id}/multi-hits and its CSV, by publication or by sequence, with the identity, coverage, date and office filters applied to each search.
  • IUPAC codes in a DNA query: when the query box holds DNA with codes such as R or N and the method reads them as letters, a note under the box says what that method does with them and offers Search as a pattern, which reads each code as the bases it stands for. The query box keeps its text, so the same primers can be searched both ways.
  • Claimed sequences, an add-on: where a server runs it, the Claimed column says whether claims of a hit's US applications and patents name its SEQ ID NO, the hit's pane gives the claims under each publication, and Claimed only keeps those hits. The CSV and Excel exports carry the same.

Sequences 0.105.0

Search engine 0.21.0

Download a completed search as an item for the Lab Notebook Inbox.

  • New: Download for Lab Notebook saves a portable item with the search record, all stored matches (including excluded matches) and file fingerprints. Import it in the notebook Inbox, add it to an entry and sign the entry. Packages are limited to 32 MiB expanded. See Exports and search records.

Sequences 0.104.0

Search engine 0.21.0

The top bar gains a Features menu and, signed out, a Start free trial button. Your account now has its own pages, with one account button on every signed-in page.

  • New: the top bar's menu opens with Features, which lists each feature page with Search methods first, then Data, Benchmarks, Examples, Pricing and Docs. The roadmap is on the docs home.
  • New: signed out, Start free trial sits beside Sign in. Like the button on the plans page, it opens the sign-in page.
  • On a phone, the top bar shows the bracket mark and Sequences on one row, and the menu opens from Menu, with Sign in and Start free trial at its top.
  • New: once you are signed in, your account button sits at the top right of every page, with your initials. It opens My work, Account, API tokens, Usage for an admin, Docs and Sign out, in place of the email address and small buttons at the top of the search page. See Your account.
  • New: Account has tabs for your profile, your plan, your API tokens, your email alerts and the searches you started over the last 7, 30 or 90 days.
  • On the hosted service, Manage your ConductScience Account on the Profile tab opens your ConductScience Account's page. Nothing links to the store's My account page any more.
  • Signed in, the docs and the site's other public pages show Search, My work, Examples and Docs in the menu.
  • The local workbench has no accounts, so it has no account button or account pages.

Sequences 0.103.0

Search engine 0.21.0

The ST.26 check has its own page, open without sign-in, two use-case pages show how Sequences fits a search, and Plans and prices answers common questions. Hit exports gain gene columns, and Data sources shows Drug names and when each source updates.

  • New: the ST.26 check has its own page, open without sign-in.
  • New: two use-case pages show how Sequences fits a search: Freedom to operate and Antibody CDRs in patents. The list of all pages has a new Use cases group.
  • New: Plans and prices answers four common questions, including what each plan includes and how the free trial works.
  • Shared links to the public pages now show a card with the page's title and description.
  • New: hit exports, the CSV from Alignments and the Excel workbook's Alignments sheet, end with three gene columns after the drug columns: genes, gene_ids and gene_links, each gene's symbol, its NCBI Gene ID and how the hit links to it, empty for a hit with no gene. See Exports and search records.
  • New: Data sources shows Drug names under What shows beside hits, where the server runs it: what it holds, its drugs, protein chains and oligonucleotide strands, the ChEMBL release and the GSRS date, and its license, each read from the data the server has loaded. Every row now says when it updates.

Sequences 0.102.0

Search engine 0.21.0

Results can now filter by Published on or after, so two dates show what came between them, and a new watch can count matches as new from a day. A site menu leads to methods, data sources, benchmarks and examples, and a new Data sources page gives each source's count and date on this server.

  • Published on or after. Results, saved views, plans, the combined search's list, the charts, the Excel workbook and the API (publishedFrom) filter by a first date as well as Published before, so the two dates show what came between them, such as the intervening art between a claim's priority and filing dates.
  • A sequence passes when none of its dates falls before your date, and a sequence with no date passes both. A new watch can count matches as new from a day (Count as new from, newFrom in the API): the matches dated on or after it show as new at once, and each weekly addition then lists what it adds.
  • New: the site menu reads Methods, Data, Benchmarks, Examples, Roadmap, Pricing and Docs, each where the server has that page. User guide, Installation guide, Release notes and All pages leave the menu, and the docs home and every footer still link to them. Signed in, the bar shows a small Search button, then My work.
  • The ConductScience logo and the word Sequences now link home from every page: to the home page before sign-in, and to the search page after. Sequences sits on the logo's baseline at every screen width.
  • New: Data sources shows what each data source holds, with its count and date on this server, how often it updates and its license. The counts and dates are read from the data this server has loaded, each time the data changes. A license is named where the source states one.
  • Choosing a search method now gives each of its seven methods a card: what it is for, what it searches, its settings, its limits, an example and its part of the user guide.
  • The home page is shorter. It starts with Before you sign in, with a card for each of Methods, Data, Benchmarks and Examples, and What Sequences does is now one list of the feature pages.
  • Fixed: on a case with claim features that no search covers, such as a fusion protein, the results summary said every claim feature was met. It now lists them among the features no reference meets, each marked "no search yet". A decision already made on such a case's results summary is asked again.

Sequences 0.101.1

Search engine 0.21.0

Fixed: where a drug's ChEMBL and GSRS records hold different chains of an antibody, as cixutumumab's and tovetumab's do, a hit that holds both variable domains now names both and credits both sources.

Sequences 0.101.0

Search engine 0.21.0

An examples page for the main parts of Sequences, a benchmarks page with each recorded run of the search methods, and My work, one signed-in page that lists your searches, imported result files, cases and watches.

  • New: the sample results page is now the examples page, with a section for search, results, tools, claims and plans, model features (Pilot) and benchmarks, each linking a working example. See Examples and sample results.
  • New: on the sample claim suggestions (Pilot), each suggestion has a link of its own and lists the checks that kept or dropped it, in the order the service runs them, with the start of its prompt's SHA-256. Each model draft (Pilot) on the sample case has a link of its own. See Examples and sample results.
  • New: Benchmarks, a new page in the site menu, shows each recorded run of the search methods: what it measured, the data it ran on, what it found and the evidence record that holds its numbers.
  • It covers exact and pattern counts checked against GNU grep, the exact scan and BLAST on short DNA fragments, finding altered sequences, Within k changes on the fragments Smith-Waterman missed, repeat runs, the ST.26 listing check, a search plan for every SEQ ID NO of one patent, and searches an hour with more workers. Three of its sections draw a chart, each above a table of the same numbers.
  • New: My work, one signed-in page, lists your searches, imported result files, cases and watches, newest first, with a filter by kind and a link back to each. 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. See My work.
  • A search of several sets at once shows as one search. Each card says what its card on the search page says, and Open goes back to it.
  • On a free trial, the page shows the days left, as the search page does.
  • The page reads only your own work and writes nothing. The local workbench has no accounts, so it has no My work page.

Sequences 0.100.0

Search engine 0.21.0

Drug names on matching hits, on for hosted searches, and two new sets to search: Drug proteins and Drug nucleic acids.

  • New: Drug names are on for hosted searches. A hit whose bases are an antisense or siRNA drug's, or whose chain or variable domain is an antibody drug's, names the drug and its highest development stage where ChEMBL gives one. Sequences compares the bases alone, not an oligonucleotide drug's chemical changes. See Drug names.
  • New: Search in offers two new sets, Drug proteins and Drug nucleic acids: each drug's chains from ChEMBL and GSRS, one record per chain.

Sequences 0.99.0

Search engine 0.21.0

Species on each hit, and a species filter.

  • New: each hit shows the species its records state, and results filter by species or by group, such as human, mouse or camelids, without a new search. The hits CSV ends with the columns species, taxa and species_groups. See Species.
  • New: on your own server, a taxa folder gives each record its NCBI taxon and group. The taxa-scan and taxa commands make it offline from NCBI's taxonomy and your snapshots' source files, and Sources and terms then credits NCBI Taxonomy. See Species of each hit.

Sequences 0.98.0

Search engine 0.21.0

On a server that runs the genes add-on, as the hosted service now does, a new search's hits on reviewed human proteins, and patent protein hits identical to them, show each gene the Swiss-Prot entry names.

  • New: on a server that runs the genes add-on, as the hosted service now does, a new search's hits on reviewed human proteins, and patent protein hits identical to them, show each gene the Swiss-Prot entry names. See Genes.
  • Changed: antibody names leave out 11 chains whose text in chembl_37 is another antibody's, letter for letter, or out of order: both chains of GANITUMAB, LUCATUMUMAB, MOGAMULIZUMAB, TOSITUMOMAB and IBRITUMOMAB TIUXETAN, and the heavy chain of CIXUTUMUMAB.
  • Changed: the pricing page now points to ConductCare for installation help.

Sequences 0.97.0

Search engine 0.21.0

Known variants, an add-on: a protein hit identical to a human Swiss-Prot entry shows the variants ClinVar records on that protein and, for a protein query, each place the query differs from it, marked when ClinVar records that change, on a server whose settings name the add-on.

  • On the Alignments tab, such a hit has two more columns, Known changes and ClinVar variants.
  • Its pane names each of the query's changes in the protein's numbering, such as p.Val600Glu, with a line for each ClinVar record of it, and a closed ClinVar variants part lists the protein's ClinVar variants, filtered by Clinical significance. See Known variants.
  • A change counts only where ClinVar describes it on a RefSeq protein identical to the Swiss-Prot entry, with the stated reference residue in place. "Not in ClinVar" means this release of ClinVar records no such change on the protein, not that the change is new. The part names its sources and their dates and draws no conclusion. Sources and terms has cards for ClinVar, RefSeq and UniProt.
  • Export matches (CSV) and the Alignments sheet of Export all tabs (Excel) end with three more columns, known_changes, query_changes and clinvar_variants, empty where known variants does not apply.
  • An add-on runs only on a server whose settings name it, and a server that names none holds none of its data.
  • On a server you run, BIOSEQUENCE_ADDONS takes variants, and BIOSEQUENCE_VARIANTS names a known variants folder from ConductScience, about 670 MB, which load checks file by file. See Known variants.

Sequences 0.96.0

Search engine 0.21.0

Query tools that turn a claim's words into searches, a sample result laid out as a result's own page, which plan fits on the plans page, the user guide's section on cases in six steps, Within k changes on the methods page, therapeutic antibody names on antibody CDR matches where the server has a name table, and a FIPS image for your own server.

  • New: Sequence tools has a Query tools tab. A claim phrase such as "at least 90% identical to SEQ ID NO: 1" becomes proposed queries, a method and a results filter, each with its reason. It also makes variants such as K26R, windows, a DNA pattern for a peptide and queries for modified residues. A query group keeps queries and searches them as one batch. See Query tools.
  • New: antibody CDR matches can name the therapeutic antibodies they share sequence with, by a whole heavy or light variable domain or a chain's three CDRs under the search's numbering, from ChEMBL under CC BY-SA 3.0. Each name links to its ChEMBL page, the CSV lists them, and a server needs a name table first. See Antibody names.
  • New: on a server with patent context, a US publication's pane shows What the patent says: which claims and paragraphs name each matching SEQ ID NO, and the abstract, those claims and paragraphs and the listing's notes, quoted with links to the full text. Compare shows up to three side by side. See What the patent says.
  • New: 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. Its line under the hit and the hits CSV show that date, marked (Crossref), and Published before compares it. When Crossref gives only a year, the paper keeps its year. See Dates from Crossref and Dates.
  • New for your own server: a FIPS image, tagged <version>-fips, for a server whose policy requires FIPS 140 validated cryptography. It runs each algorithm the service uses in OpenSSL's FIPS provider, which holds a FIPS 140-3 certificate, apart from the license file's Ed25519 signature, and its fips-check command shows where each one ran. See Load the image.
  • Changed: a sample result's page now has the layout of a result's own page, 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. See A sample result.
  • Changed: the user guide's section on cases opens with its six steps, each linked to its part, then follows one claim through them, with tables and drawings in place of long paragraphs and the fine print in folds that open in place. The methods page adds Within k changes and its runs on the benchmark. See Cases and Within k changes.
  • Changed: the plans page says which plan fits six kinds of buyer, from patent attorneys to biotech companies, universities and investors, and how to set up your own server: from the installation guide, with no setup fee, or with our engineers by the hour. On the hosted service, the home page shows the four plans and their prices, and four feature pages link demo data to try each one with.

Sequences 0.95.2

Search engine 0.21.0

0.95.1 with its test image's build fixed.

  • Fixed: the image's test stage builds again. In 0.95.1 it stopped on a script the build was not given. The service image was not affected.

Sequences 0.95.1

Search engine 0.21.0

The features of 0.95.0, with one fix to which match a publication's pane names when two match equally well.

  • Fixed: when two of a publication's sequences match your query equally well, its pane and the line on why it is listed name the one with the lowest SEQ ID NO, then the first by name. Before, the one they named could change when the stored sequences were updated, though the hits were the same.

Sequences 0.95.0

Search engine 0.21.0

Within k changes, why each publication is listed, each result on its own page, search settings, repeatable Smith-Waterman and global searches, result files matched by their columns, saved views and transformations, demo data for every input, and public pages for visitors.

  • New: Within k changes lists every stored sequence within k changes of a query, up to 5 substitutions or 5 changes counting insertions and deletions. The counts are complete, and each query lists up to 50,000 hits, the closest first. An N, an X or another IUPAC code in a query matches any letter it stands for, and stored DNA is read letter for letter. See Within k changes.
  • New: a publication's pane on Patents, and its open row in a case's References step, opens with why it is listed: how its sequences match your query, where it falls against your date, and the earliest publication holding your query unchanged. An X in a stored protein, or an N in DNA, is marked apart from a mismatch, and a filter can count it as a match. See By patent.
  • New: each result has a page of its own, with the search's name in the browser tab, so you can bookmark it and come back. Edit and run again fills the search form with the search as it ran, to change and run as a new search. Older result links and the links in watch emails open the new page. See Addresses.
  • New: Search settings sets the scoring matrix, gap costs, word size, low-complexity filter and genetic code for Alignment (BLAST) and the FASTA36 methods, each with its default and range, and the search record lists them. In the Sequences image, Smith-Waterman and global searches now return the same hits and E-values every time they run. See Search settings.
  • New: a case's Report step imports any comma- or tab-separated result file with a header row. Before it uploads, you match its columns to the query, hit, publication, SEQ ID NO, identity, coverage and score. Other columns stay with each row, and an import report says what was read and skipped, and why. Use a demo CSV to try it. See Result files from other programs.
  • New: Views saves the way you read a result, its tab, filters, sorts and columns, under a name kept with the search. Transform folds hits with the same alignment and keeps the hits that cover or differ at chosen positions, each change undone with Undo. More downloads saves the hits as FASTA, the alignments as text and the kept patents as a reference list. See Saved views.
  • New: on the hosted service, a visitor who is not signed in sees a home page with a page for each main feature, a security page and an accessibility statement, and every server adds an API reference. Use demo data beside each input fills it from a published patent or a public database entry, and the sample results start with a sample case, shown read only. See Demo data.
  • Changed: the plans page lists Enterprise and a deployment in your own cloud beside the hosted service and the server edition, with the hosted service's free 30-day trial and the server edition's free 90-day evaluation. The methods page adds One query, seven methods, the same query run by each method side by side. See Choosing a search method.

Sequences 0.94.0

Search engine 0.21.0

Results have charts, chosen columns, groups, a comparison of searches and an Excel export, every public page has one menu, and two new pages explain a result and the search methods.

  • New: a result draws three charts above its tabs: identity against coverage, hits by year against your date, and where the first 20 hits align on your query. Each follows the list's filters, shows its numbers as a table and saves as SVG or PNG. Drag across the first chart, or use the keyboard, to keep only the hits in that area. See Charts.
  • New: Columns on Patents and Alignments shows, hides and orders the columns, and keeps your choice for your account. Group can list patents under a heading for each applicant, patent office or year. Export all tabs (Excel) saves every list and the search record in one workbook. See Columns.
  • New: Compare searches, after your recent jobs, shows which hits two to six finished searches share and which only one found, each with its measure in every search. The sample results show the same charts, columns and groups, and download as Excel. See Compare searches.
  • New: every public page has the same menu at the top, with the search, the guides, the release notes and a list of all pages. On a phone or tablet it opens from Menu. The docs home lists each section of the user guide. On the hosted service, search engines may now list the sample results too. On a server you run, they stay unlisted.
  • New: How to read a result walks through one hit of the GLP-1 sample: its alignment, drawn letter by letter, its four measures worked out, and its place against a date. Choosing a search method says which method fits which query, with a drawing of each and the recorded runs behind the choice. The user guide links to both.
  • Changed: the PDF case report embeds its fonts and carries its metadata, and veraPDF's PDF/UA-1 check finds no failure in it. Watch emails have a title and a heading. Pages are now tested in Firefox as well as Chromium, and at 400% zoom, and at 320 px the results list keeps publication numbers readable.

Sequences 0.93.0

Search engine 0.21.0

A document system can send an application into a case, the roadmap gives each planned item a day, and model suggestions and drafts use new prompts.

  • New: a document system can send an application's claims, sequence listing and publication XML into a case in one request, with a connector token an admin makes for named accounts on the API tokens page. Sending the same set again changes nothing, and a reference program sends each folder put in an inbox. See Connectors.
  • New: the installation guide's searches an hour add four steady loads on 12 CPUs, each keeping 4 searches queued or running for 30 minutes: proteins and DNA by the automatic method, exact text and patterns, each with its median and 95th-percentile time, CPUs busy and most memory. See Searches an hour.
  • Changed: the roadmap gives the day each planned item is expected and lists ten new items, among them Smith-Waterman and global searches that return the same hits every time, a search for every stored sequence within a set number of changes, and fast, complete search for short DNA fragments such as primers and probes.
  • Changed: claim suggestions from a model use prompt 3. It names each suggestion's claim by its number in the claims list and quotes that claim's own text, not another claim or a paragraph of the description, the two commonest reasons a suggestion was dropped. See Claim suggestions from a model.
  • Changed: a model drafts the results summary with prompt 3 and the report sentences with prompt 2. The results prompt gives each count the words the results template uses, and the report sentences prompt asks for the facts without judging the publication. Drafts made before and the decisions on them still hold. See Summary drafts.
  • Fixed: on the container image, a call to the claims model that took over 5 minutes failed with "The claims model did not answer.", whatever BIOSEQUENCE_CLAIMS_MODEL_SECONDS said. Now such a call, for claim suggestions or a drafted summary, waits as long as that setting allows: 900 seconds by default in the openai-chat format. See Claim suggestions from a model.

Sequences 0.92.1

Search engine 0.21.0

The hosted service lists its public pages for search engines. Nothing changes for a server you run.

  • Changed: on the hosted service, /robots.txt and /sitemap.xml open without signing in, so search engines can find the docs pages, the release notes, the roadmap, the list of all pages and the plans page. A server you run answers neither.

Sequences 0.92.0

Search engine 0.21.0

The hosted service counts visits to its public pages with Google Analytics. A server you run loads no analytics.

  • Changed: on the hosted service, the docs pages, the release notes, the roadmap, the list of all pages and the plans page load Google Analytics 4, which counts their visits. The search page, sign-in, the sample results and every page after you sign in load none, so analytics never sees a search, a case or an account. A server you run loads no analytics.

Sequences 0.91.2

Search engine 0.21.0

The roadmap lists the summaries a language model drafts as released, in pilot.

  • Changed: on the roadmap, Summaries drafted by a language model your server runs moves from Next to In pilot now: released in 0.89.0 and marked (Pilot). A draft enters a case's report only when you accept it or edit it.

Sequences 0.91.1

Search engine 0.21.0

0.91.0 with a test fixed so that it installs, and TrEMBL and ESTs in the installation guide's sizing tables.

  • Fixed: one of 0.91.0's tests renamed a file in a snapshot's folder, which is read-only, and Linux refuses that, so 0.91.0's install stopped at its tests and the hosted service went from 0.90.0 to 0.91.1. The service's code is 0.91.0's.
  • New: the installation guide's measured searches add TrEMBL and ESTs, three for each method, so every set the hosted service offers is measured. See Measured searches.

Sequences 0.91.0

Search engine 0.21.0

A case keeps its logs whole after a stop and decides on the words it shows, watches that cannot run say why, the search page and a case's screens wait out an ended sign-in, and a record that cannot be read no longer stops the start.

  • Fixed: after a stop between an event and the record that names it, the next event in a case's log, the audit log or an account's join log took a number already used, and Verify failed from then on. The next event now follows the log's last line, and the log verifies again. A stop part-way through locking a plan no longer leaves a version a run can name.
  • Fixed: in a case's Review step, accepting a finding already accepted, or rejecting one already rejected with the same note, logs nothing more, and an open form keeps what you typed when a model's suggestions arrive. Accepting a rejected finding drops the rejection's note. The Plan step starts afresh in each case, and Unlock the plan asks for its reason first on a plan behind the current claims.
  • Changed: a decision under Why each reference counts is on the words the page shows. Each sentence of GET /api/cases/{id}/explanations has fingerprint, which POST takes back as ifFingerprint; a sentence that reads differently now answers 409 with the code changed-elsewhere, and the page shows its new words. See Why each reference counts.
  • Fixed: once an account's free trial has ended, its watches no longer run, and each says so with no Resume. A watch paused for its size offers Delete and no Resume, and POST /api/watches/{id}/resume on it answers 409 with the code not-allowed-in-state. A watch whose snapshot comes back loses its line at the next hourly check. See Watches.
  • Fixed: start.mjs drain no longer waits without end. It exits 1 when the worker starts again before it drained, and after 60 minutes or the minutes drain --wait gives, from 1 to 1440, and the drain stays asked. The Servers section names Starting and Waiting for the coordinator in words. See Drain and stop a worker.
  • Fixed: the case opened last is the one that shows, and a case the server no longer has shows the list of cases with "This case is not on the server." Once your sign-in has ended, the page asks the server nothing on a timer and a case's steps say so in one sentence; each check goes on by itself after you sign in again. See When your sign-in ends.
  • Fixed: the result you opened last is the one that shows, and Back or Forward to a search no longer on the server closes the open result. A results list that does not load says so in place of its Loading line, with Try again. K or X pressed twice before the list redraws marks the row once, and a family's mark that stops part-way shows what was saved. See Results.
  • Fixed: a search's job.json, calibration.json or a database's manifest.json that cannot be read no longer stops the service from starting or empties a list. The service skips it and leaves it on disk, names it in the start log and counts it in /healthz as unreadable. See Health and logs.

Sequences 0.90.0

Search engine 0.21.0

The search page names a search's state the same way everywhere, offers Run again where a search ended, and says when the service is not answering.

  • Changed: 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 with a set that did not complete, whose result names each such set with Run again for it. See While a search runs.
  • New: Run again on the card and in the status bar of a search that failed, was cancelled or was interrupted. Run search pressed again while the same search waits or runs starts none and follows it, and POST /api/jobs/{id}/again on such a search answers 409 with the code not-allowed-in-state. See Run again.
  • New: 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 says so, with Close.
  • Fixed: the Search record tab no longer shows Export, which saved the list open before it; a search whose timing cannot be saved completes; an import that does not complete leaves no folder; and each footer link is a target at least 44 px wide as well as tall.

Sequences 0.89.0

Search engine 0.21.0

Summaries drafted by a model the server names, in pilot, and plan runs that can be cancelled.

  • New (Pilot): on a server whose operator names a model, an examiner can ask it to draft the case summary, the results summary, a claimed SEQ ID NO's summary or a kept publication's report sentences. A draft whose quotes or numbers are not in the case's facts is dropped, and only a draft the examiner accepts or edits reaches the report. See Drafts from a model.
  • The case log keeps each draft and decision, with the model, the prompt and the tokens of each call. Drafts share the model with claim suggestions, one call at a time. The hosted service names no model, so drafts are off there. See Summary drafts.
  • New: a plan run that waits or runs can be cancelled, with Cancel this run on the case page or POST /api/cases/{id}/runs/{runId}/cancel. Its searches that wait or run stop, the rest do not start, and finished searches keep their results. GET /api/cases/{id}/runs/{runId} reads one run.
  • Fixed: cancelled and interrupted searches count apart from failed ones; a second Run click during a run starts none; case cards and heads say searches running or ran; an admin's delete cancels the case's runs first; the page stops asking for a model's answer after 3 failed reads; and a cut-off last line in a case's file is skipped.

Sequences 0.88.0

Search engine 0.21.0

The AWS install kit is no longer a beta, since an install in an AWS account passed every check; a published way to report a vulnerability; a roadmap that leads with what comes next; the ConductScience logo on case reports; and guides that give measured times in place of caveats.

  • Changed: the AWS install kit 1.1.0 is no longer a beta: an install in an AWS account passed every check on November 14. Deleting its stack now keeps the data volume, and the guide's removal steps say how to delete it. See In your AWS account.
  • New: Report a vulnerability names security@conductscience.com and how soon a fix comes: within 7 days for a critical vulnerability and 30 days for a high one. Every server answers /.well-known/security.txt with that address.
  • Changed: each release's image is scanned before it is installed on the hosted service, and a high or critical finding with no recorded reason stops the install.
  • New: the self-host control list covers every control and enhancement of the NIST SP 800-53 moderate baseline, and a secure development page maps Sequences to NIST's SSDF (SP 800-218), each gap stated.
  • Changed: a server whose health check fails three times in a row restarts by itself, and two workers on one host share a network of their own, so each joins again after the coordinator restarts.
  • Changed: the roadmap opens with Next, then In pilot now, then Recently released: the ten newest, the rest under a disclosure, and a link to all release notes.
  • New: PDF and Word case reports open with the ConductScience logo beside the line that says who made them.
  • Changed: the hosted sign-in names Sequences, with no beta label; a link from another site opens the local workbench; sample pages leave out a search's full limits; the guides give measured times in place of caveats; and after a trial, /pricing and the service say to write to support@conductscience.com to choose a plan.

Sequences 0.87.4

Search engine 0.21.0

The installation guide's sizing tables: searches measured on the hosted service on its patent sets, Swiss-Prot and NCBI core_nt, with their CPU time, run time, memory and data read, its searches an hour in a load test, the size of its snapshots on disk, and the AWS prices of the kit's server types.

  • New: Sizing and cost in the installation guide gives three searches measured on the hosted service for each method on Patent DNA, Patent proteins and Swiss-Prot, and three BLAST searches of NCBI core_nt, with their CPU time, run time, peak memory and data read, and the average CPU time of a search by method and set. See Measured searches.
  • New: the same section gives the searches an hour of a load test, the size on disk of each snapshot and of NCBI core_nt, and the on-demand AWS prices of the server types the AWS kit offers. See Sizing and cost.

Sequences 0.87.3

Search engine 0.21.0

The same service as 0.87.2, with the new ConductScience icon in the browser tab.

  • Changed: the browser tab and a phone's home screen show the new ConductScience icon, which browsers fetch again in place of the one they kept.

Sequences 0.87.2

Search engine 0.21.0

The same service as 0.87.1, with the new ConductScience logo on every page.

  • Changed: every page's top bar shows the new ConductScience logo, and the browser tab shows its mark.
  • Changed: the roadmap lists several servers sharing one queue as released in 0.81.0. See Several servers.

Sequences 0.87.1

Search engine 0.21.0

The same service as 0.87.0, with one fix. The hosted service took the compute for each search and compute limits with this release.

  • Fixed: a search whose set file cannot be read fails with its program's own message again, as before 0.87.0, not with one about the size of the file. The 0.87.0 notes list what both releases add.

Sequences 0.87.0

Search engine 0.21.0

Compute for each search: every search records its CPU time, run time, peak memory and the data it read, the usage page adds them up by account, day and set with a CSV, and an admin can cap each account's CPU hours a day and have long or large searches ask before they run.

  • New: each search records the CPU time of the programs it ran, its run time, their peak memory and the size of the set files they read, in its job and its search record. See What each search uses.
  • New: the usage page shows CPU time and data read for each account, and CPU time, run time, peak memory and data read for each day by account and by set. Download CSV saves one row for each day, account and set. See Usage.
  • New: an admin can cap the CPU hours each account's searches use in a UTC day. At the limit, new searches wait until 00:00 UTC, and the message says how long that is. See Compute limits.
  • New: a search of a set the admin names, such as TrEMBL, or one estimated to run past the time the admin sets, asks before it runs, with its estimate. See Searches that ask first.

Sequences 0.86.2

Search engine 0.21.0

A fix to follow-up searches in a case: a suggestion you added or dismissed stays that way when a later plan version runs the same search again.

  • Fixed: a suggestion you added or dismissed came back when a later plan version ran the same search again. It now shows the plan version and search that hold it, or stays hidden, so Add to plan cannot put a search in the plan twice. See Follow-up searches.

Sequences 0.86.1

Search engine 0.21.0

Fixes to cases: the earliest priority date from a US provisional application, claim counts, run progress in search jobs, imported results under their claim features, and follow-up searches in a set kept in another snapshot.

  • Fixed: a publication that claims a US provisional application gives that date as its earliest priority date, so the Claims step offers it as the prior art date. See Cases.
  • Fixed: cancelled claims written as one entry, such as "10-11. (canceled)", count as each claim they stand for, so claims 1 to 24 are 24 claims wherever a case counts them.
  • Changed: a run's progress line and the case summary count search jobs, as a search of many queries runs as several: "7 of 12 search jobs done, about 4 minutes left."
  • Fixed: a decision on the case summary stays current while other searches on the server finish. Once a plan runs, the summary no longer depends on its expected time, which changes each time the server times a search.
  • Fixed: a result file you import whose queries are named SEQ_ID_NO_<n>, as the search request names them, puts its references under the claim feature that recites each SEQ ID NO, as a plan search does. See Result files from other programs.
  • Fixed: Set not searched, under a claim feature in the References step, offers a set kept in another snapshot, such as Swiss-Prot beside a snapshot of the patent sets alone. See The References step.

Sequences 0.86.0

Search engine 0.21.0

A summary at the top of each case, a summary of a case's results, and All references, which lists each reference once across the case.

  • New: Case summary, under the case header, opens to one paragraph about the whole case: the application and prior art date, the claims and their sets, the features, the listing and its WIPO ST.26 check, the claimed SEQ ID NOs, the plan and where the case stands. Sequences fills it in from a fixed template, with no language model. See Cases.
  • New: Summary of the results opens the References step once a search adds references. It counts the references, each once, says how many meet a claim feature's threshold and came before the prior art date, and names the first 5 and the searches they come from. See The References step.
  • Accept, edit or reject each one, as with the other summaries. The report prints an accepted case summary first, and the summary of the results first under Summaries. See Summaries.
  • New: 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 features it meets. Open a row to see its rank, match and mark under each feature. See All references.
  • Download CSV under All references saves each reference once, with its rank, best match and mark under each feature. In the API: GET /api/cases/{id}/references/all and /references/all.csv, and summaries of kind case and results.

Sequences 0.85.0

Search engine 0.21.0

Follow-up searches suggested from each result, a warning before your sign-in ends, a list of all pages, and regular expressions in pattern search.

  • New: Follow-up searches under a finished result suggests up to three searches to run next, such as a more sensitive method when a search finds few matches, more hits when it reached its limit, or the part of the query its best match leaves out. Each says why, from the result's own numbers. See Follow-up searches.
  • Run starts a suggested search, Apply filter narrows the list and Dismiss hides one. In a case, Add to plan puts a suggestion in the draft plan with its rule as the reason, and the case log records each one shown, added or dismissed. See The References step.
  • Five minutes before your sign-in ends, a message says when. Choose Stay signed in to sign in again in a new tab. Once it has ended, the page no longer reloads when you act: a message offers Sign in again. Either way, what you typed stays on the page. See When your sign-in ends.
  • Every page's footer links to All pages, a list of the pages on this site. An address the site does not know opens a page that says so, with a link to the search page.
  • New: Pattern syntax above the pattern box picks PROSITE or Regular expression. A regular expression takes letters, [AG], [^P], ., {2,6}, ?, +, *, ^, $ and up to 8 choices such as (GGATCC|AAGCTT), and runs as the PROSITE patterns it stands for, with the same limits on both strands. See Exact and pattern search.
  • The search record keeps the expression as you typed it and its PROSITE form, and Run again searches the same way. A backreference, a lookahead or lookbehind, or a repeat on a group such as (AG){3} is refused with a message that names the part. See Run again.
  • In a PROSITE pattern, < before the first part ties a match to the start of the stored sequence and > after the last part to its end. In the API, a pattern search takes patternSyntax: prosite, the default, or regex.
  • Fixed: a sample result made from a pattern search shows the pattern as its query, such as MKT-x(2)-IAK, named Pattern or Regular expression. Remove and add again any sample made before this release to correct it. See Sample results.

Sequences 0.84.0

Search engine 0.20.0

Signing out now deletes what that browser was still uploading or exporting, each search deletes its working files when it ends, and administrators can set how long records are kept.

  • Signing out stops the uploads and exports that browser was still making and deletes their files, and the page clears what it kept in the browser. A sign-in that ends on its own does the same within a minute. Your searches, listings and cases stay, and your other browsers keep their own work. See Who can see your work.
  • Each search and alignment deletes the files it made on the way, such as its query in parts and the engine's raw output, once it finishes, fails or is cancelled. Its results and record stay, and Run again works as before. See Working files.
  • New setting BIOSEQUENCE_RETENTION_DAYS: set to a number of days, it removes searches that no case holds, portfolio exports, and combined searches and batches once they are that old. A search a case holds or a watch follows stays. Unset, every record stays. See How long records are kept.
  • Administrators can delete a case with its searches through the API (DELETE /api/admin/cases/{id}). A search still running is cancelled first, and listings stay. The audit log keeps the case's ID, its owner and the SHA-256 of its log's last event.
  • The audit log notes each deletion with counts, never a file's name or content: files.deleted, session.ended, records.removed and case.deleted. See Audit export for a SIEM.
  • Time estimates for Smith-Waterman on DNA now follow times measured on the current patent DNA set. A short query that followed a long one was estimated about a quarter too short.
  • The first start after this upgrade deletes the working files that earlier releases left in search folders. Every search keeps its record, request and results.

Sequences 0.83.0

Search engine 0.20.0

A case's plan groups its SEQ ID NOs, shows how long it should take, splits a large plan into runs that start one after another, and saves a search request as Word or JSON.

  • New: Query groups in a case's Plan step groups the claimed SEQ ID NOs by molecule type, length, identity and method, with the claims each group serves and the searches and jobs it becomes. Move SEQ ID NOs, Split and Join another group change them before the lock. See Query groups and large plans.
  • New: Size and time gives a plan's queries, letters, searches and jobs against this server's limits, with the time it should take. A plan of more than 100 search jobs runs as several runs, each starting on its own when the one before has finished.
  • New: from a locked version, Download search request (Word) and Download search request (JSON) save each group's sequences in FASTA with its method, sets, thresholds and reasons, for a search team or another system. The case log records each file's SHA-256.

Sequences 0.82.0

Search engine 0.20.0

Administrators can define export formats with their own columns, file types and redaction rules, and every export now records its format and the file's SHA-256.

  • Your administrator can define export formats in a JSON file for matches, patents, a combined search's references and a case's references: each with a name, a version, an audience (internal or external), its columns and headers, CSV, TSV or JSON lines, and fields to leave out or mask. See Export formats.
  • When formats are set up, a Format choice shows beside the export button. Built-in (CSV) saves the file it always did. After a download in another format, a message says how many rows the file holds and what it left out. See Exports and search records.
  • A new setting names the sets that hold confidential data, such as unpublished applications you attach. Every external format leaves out each row such a set found, and no setting turns that off. See Confidential sets.
  • Each export is noted with its format's name, version and SHA-256, the file's SHA-256, its rows and what it left out, never what the file holds: a search's exports at GET /api/jobs/{id}/exports, a combined search's in its record, and a case's in its log.
  • The service checks the formats file at start and stops with a message that names the format and the problem. A format changed without a new version also stops the start, so each record names the format that wrote its file.
  • Each download of a case's references adds a line to the case log, and the Log tab counts it at once. See The References step.

Sequences 0.81.0

Search engine 0.20.0

One install can now run on several servers sharing one queue: a coordinator holds the pages, sign-in and results, and workers run its searches. Servers without the new settings run as before.

  • BIOSEQUENCE_ROLE sets each server's part: single, the default, coordinator or worker. A coordinator needs sign-in on and the worker key, set with BIOSEQUENCE_WORKER_KEY_FILE: one random value of 32 characters or more, the same on every server. Workers reach the coordinator on its worker port, 8766 by default. See Several servers.
  • A worker names its coordinator with BIOSEQUENCE_COORDINATOR_URL and itself with BIOSEQUENCE_WORKER_NAME. It holds the same snapshots, deltas and paper folder in its own data volume, and needs no license file, sign-in or published port. For workers on other hosts, BIOSEQUENCE_WORKER_BIND, BIOSEQUENCE_WORKER_TLS_CERT and BIOSEQUENCE_WORKER_TLS_KEY open the worker port to them over TLS.
  • Before a worker takes a search, the coordinator checks that it runs the same release, search programs and paper folder, and that its copy of each snapshot and delta has the same SHA-256 list. The coordinator checks every result before the search completes. load keeps the list it checked each file against, so a copy it loaded is ready at once.
  • On a coordinator, BIOSEQUENCE_CONCURRENCY takes 0 to 16, and at 0 the workers run every search of a snapshot. Searches of attached databases, Align exactly, multiple alignments, antibody numbering, portfolio exports and imported result files stay on the coordinator. The time a combined search or a case run has left counts every search that runs at once.
  • drain, run in a worker with docker exec, lets it finish its searches and take no more, then ends with 0, so the worker stops with nothing cut off. A search whose worker stops or stops answering waits in the queue again, in its place. One cut off three times fails with "The search was cut off 3 times before it finished. Submit it again." See Drain and stop a worker.
  • A coordinator keeps its searches of a snapshot through a restart: one that waits keeps its place, and one it ran itself starts again. A search a worker runs goes on if the worker reaches the coordinator within a minute of the coordinator's start.
  • On a coordinator, the usage page ends with a Servers section for admins: each worker's state, the searches it runs, how many it runs at a time, the copies that match and its last contact. Programs read the same list from GET /api/admin/workers. The OpenAPI description now gives it, the job's shared, worker, retries and ahead, and the status's servers.
  • The coordinator's audit log notes each worker that joins, leaves, is lost or is refused, and each start's settings, with the worker key shown as set. The installation guide's new section covers two workers on one server and workers on other hosts.

Sequences 0.79.0

Search engine 0.20.0

The roadmap adds summaries drafted by a language model your own server runs, expected in December 2026.

  • The roadmap lists it under 2026 Q4. On a server whose settings name a language model, a case will be able to ask it to draft the summaries that fixed templates write today, and no draft counts until you accept or edit it. The hosted service runs no model.

Sequences 0.78.1

Search engine 0.20.0

A listing card's facts and a claim's start in the claims index show in full on a phone, and the docs pages link to the sample results.

  • A listing card's line of facts wraps instead of stopping at two lines, at every width. Below 768 px, on a phone or at 200% zoom, the start of each claim in the claims index wraps whole too; on a wider screen it stays a two-line preview, and the whole claim is under Review.
  • Where a server has a sample result, the row of pages at the top of each docs page now ends with Sample results, the same samples the sign-in page's "See a sample result" opens. See A sample result.

Sequences 0.77.0

Search engine 0.20.0

Keyboard, focus and screen reader fixes across the pages, tested against WCAG 2.1 Level A and AA.

  • New: Keyboard and screen readers in the user guide lists the keys, the forms Esc closes, the deletes that ask first, the status lines a screen reader reads and how long a message stays.
  • Code blocks, tables, alignments and the search record take the focus, so they scroll from the keyboard. Buttons that repeat down a page say what they act on, such as "Open results of search S1", and a short count such as 1.2M is read as the full number.
  • Deleting a listing, a saved sequence or a watch, revoking an API token, removing a search from a draft plan and withdrawing a note now ask first, since none of them can be undone. Afterwards the focus goes to the list's tab or heading.
  • Esc closes each form that opens in place in a case, Hand over among them, as Cancel does, and the focus goes back to the button that opened it.
  • A message stays while the pointer rests on it or its close button has the focus. The usage and API tokens pages link to each other. Count chips beside a case step's headings are darker, and dates, row names and buttons wrap instead of being cut off when you widen text spacing.

Sequences 0.75.0

Search engine 0.20.0

A case's owner can hand the case, with everything in it, to a colleague's account.

  • New: hand a case to a colleague. In the case header, choose Hand over, give the account your colleague signs in with, and confirm. The case goes with everything in it: claims, plan, searches and their results, imported files, references, decisions, summaries, notes and log. It leaves your list, and your colleague can hand it back. See Hand a case to a colleague.
  • 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, the account must be at your own email domain. Your other searches and listings stay yours.
  • The case log and the server's audit log record who handed the case to whom. The case header and the case's card in your list name the last handover, such as "Handed over by a@example.org to b@example.org on 13 Nov 2026."
  • A case waits while the claims model reads it or its listing imports. Handing a case over starts no searches, so it still works after a license or trial ends.
  • For programs: POST /api/cases/:id/handover with {"to": "b@example.org"} hands a case over, with an API token too. The OpenAPI description gives it as handOverCase.

Sequences 0.73.0

Search engine 0.20.0

Cases read the changes a claim allows, such as "one, two or three amino acid substitutions", and the letters a claim reads as any residue, and rank and filter matches by them.

  • New: the case's rules find two more kinds of claim feature. Allowed changes reads a bounded count, such as "with one, two or three amino acid addition, deletion and/or substitution" or "up to 3 conservative substitutions", linked to the SEQ ID NO it changes. Letters read as any residue reads a claim's own definition, such as "wherein X is any amino acid and Z is any amino acid".
  • An open count ("one or more substitutions"), a lower bound ("at least three differences") and a change at a named position give no feature. In Review you can add or edit either kind by hand, as any other feature. See Cases: claims to search plan.
  • References: a match meets a claim that allows changes when its changes of the kinds the claim names, such as substitutions, number that many or fewer and it makes no change of another kind. A claim of "mutations" or "changes" counts every kind, and query letters outside the alignment count as deletions. Where the claim gives an identity too, either one meets it.
  • The Meets column reads "More changes" for a match with more, and "Other changes" for one with a kind the claim does not allow. The words count each kind: "Meets up to 3 substitutions: 1 substitution, 93.3% identity over the full query". Sequences does not check whether a substitution is conservative, and says so: "2 substitutions, not checked as conservative".
  • A letter a claim reads as any residue matches any residue in that claim's references, so a Z the claim defines as any amino acid no longer counts as E or Q there. See The References step.
  • Plan: a search for a SEQ ID NO whose claims allow changes keeps the identity those changes leave, and its reason says so, such as "Keeps 80% identity or more, which 3 changes leave in the 15 letters of each." Sentence 2 of "Why each reference counts" says it for the claim, with the kinds of change the claim allows.
  • Cases that exist keep their features. The rules find the new kinds in a set of claims you import or make current from now on, and you can add them by hand. A summary's facts change only for a feature of either kind, so every other decision holds.
  • For programs: POST /api/cases/{id}/features takes changes and wildcard, a reference group's threshold gives changes and wildcards, and a reference's best match gives its changes and changesByKind. The decisions file gives each new feature's details, such as "Up to 3 substitutions only". The OpenAPI description gives every field.

Sequences 0.72.1

Search engine 0.20.0

A fix to cases: a query's X reads as any amino acid, as a sequence listing defines it, and a drafted plan's identity filter keeps the matches its claims cover.

  • Fixed: in a case's references, a match reads the query's letters as a sequence listing defines them. An X in a protein stands for any amino acid, so a reference that differs from a SEQ ID NO only at an X meets the claim as written: "Meets as written: 93.3% identity over the full query, and 1 position is X, read as any amino acid". See The References step.
  • The same goes for B, Z and J in a protein, and for N and the other IUPAC codes in DNA and RNA, such as R for A or G: each covers the residues WIPO ST.26 gives it, so Z covers E and Q. The search's own identity and Query Match do not change.
  • Fixed: a drafted plan's identity filter could hide matches its claims cover, since a search counts a query's X as a difference. A search for a SEQ ID NO that holds such letters now keeps less identity, by what those letters take away, and its reason says why, such as "Keeps 70% identity or more, since identity counts 3 of the 15 letters of each, X and Z, as differences."
  • A locked plan keeps its filters. To take the new rule, unlock the plan with a reason and draft it again.
  • Sentence 2 of "Why each reference counts" now says why such a search used less identity than its claims recite, such as "Claim 1 recites 90% identity or more; this search used 78%, since identity counts 1 of the query's 9 letters, X, as a difference." See Why each reference counts.
  • Where a best match now reads differently, because its query holds such a letter, the reference's summary has new facts, so it asks you to decide again. Every other summary keeps its facts and your decision.
  • Fixed: while an unlocked plan still held the searches of an older set of claims, the plan listed a fault for each search that pointed past the current claims' words. It now lists only "Draft the plan again: it was drafted for set 1, and set 2 is current.", since drafting again mends those faults too.

Sequences 0.72.0

Search engine 0.20.0

Patent DNA and Patent proteins now hold US published applications, from the first in March 2001 to this week, with US patents from the same weekly files.

  • New: Patent DNA and Patent proteins hold US published applications, from the first, published 15 March 2001, to 12 November 2026: 288 thousand applications with sequence listings and 219.6 million SEQ IDs, 172.1 million of them DNA or RNA and 47.5 million protein. They come from the US patent office's weekly full-text files and published sequence listings. See Sets.
  • US patents granted since 2001 come from the same weekly files, so a new grant does not wait months for GenBank. An application and the patent granted from it share one sequence line when their letters match, so one hit lists both, each with its own dates.
  • Patent DNA now holds 74.9 million distinct sequences, up from 28.8 million, and Patent proteins 17.7 million, up from 10.5 million, from 793 thousand publications. A search you ran before keeps the snapshot it ran on.
  • A hit on an application shows its number, kind code, publication date, applicant and title, with its filing date and earliest priority date beside them. With Published before set, a sequence first published after your date still passes when a US application holding it was filed, or claims priority, before your date. See Dates.
  • Each week's applications and grants arrive as an addition to the snapshot: choose Main plus New under Collection to search them.
  • Smith-Waterman and global alignment read every stored sequence, so a DNA search of the larger Patent DNA takes longer: a 21-base Smith-Waterman query took 54 seconds on the hosted service. See Search methods.
  • Sequence listings for 195 US applications named in the weekly files are not in the sets yet. For 142, the sequence listing site sent a file our reader could not read. For 38, from 13 January and 7 July 2022, the site returned an error each time we asked. For the other 15, no list we read from the site names a file. Each weekly addition asks again for the 38. Corrected 17 November 2026.

Sequences 0.71.0

Search engine 0.20.0

For a server you run yourself, a new guide page shows which NIST SP 800-53 controls of the moderate baseline Sequences meets, which your environment meets, and which the two share.

  • New for a server you run yourself: NIST SP 800-53 controls, a page of the guide Run Sequences on your own server. It takes each of the 146 controls and enhancements that the SP 800-53 Rev. 5 moderate baseline selects in access control, audit, configuration management, identification and authentication, system and communications protection, and system integrity.
  • Each row says whether Sequences meets the control, your environment meets it, or the two share it, and names the feature, setting, file or guide section that does. The page lists the baseline's source, NIST's OSCAL files with their commit and SHA-256 hashes, so you can check the list.
  • On a server you run, protection at rest comes from your environment, such as an encrypted volume. Each file it writes is readable only by the service's account.

Sequences 0.70.0

Search engine 0.20.0

Smith-Waterman and global alignment search DNA too, on both strands, and the status bar counts the queries done and gives the time left.

  • New: Smith-Waterman, Global on the query and Global on both sequences search a DNA set with a DNA query, on both strands, as they do protein. In our benchmark on Patent DNA, Smith-Waterman found 16 of 20 DNA fragments of 21 to 50 bases at 80% identity, where BLAST found 8. Automatic still runs BLAST for DNA. See Search methods.
  • Each DNA query reads every stored sequence, so a search takes minutes: on Patent DNA a Smith-Waterman query took about a minute at 21 bases and 5 minutes at 308 bases. The status bar counts the queries done and gives the time left in whole minutes, as in "2 of 5 queries done, about 6 minutes left". See While a search runs.
  • 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. POST /api/jobs and POST /api/estimate answer 400 with the same words.
  • A case's plan offers Smith-Waterman and global alignment for a DNA set too. For programs, a running Smith-Waterman or global search reports progress: the queries done, of all of them, and the part of the estimated work they hold.

Sequences 0.69.0

Search engine 0.20.0

A Roadmap page lists the work planned for Sequences, by quarter.

  • New: the Roadmap page lists the work planned for Sequences by quarter, with what each release since 0.38.0 added. It opens from the row of pages at the top of the documentation and from the documentation home, without sign-in.
  • Expected in December 2026: US published applications in the patent sets, added each week, and Smith-Waterman and global alignment for a DNA query. Expected in March 2027: result files from other sequence search systems in a case, and several servers sharing one queue. Plans can change; these notes record what each release added.

Sequences 0.68.0

Search engine 0.20.0

The query box, Open a file, a dropped file and the search API take PIR (NBRF) records as well as FASTA.

  • New: paste or drop PIR records, such as >P1;CRAB_ANAPL, wherever FASTA works: the query box, Open a file, a file dropped on the search card, POST /api/jobs and POST /api/batches. See Start a search.
  • A PIR header names the sequence after its semicolon, and its two letters give the type: P1 and F1 are proteins, and DL, DC, RL, RC, N1 and N3 are DNA or RNA. The line after the header describes the sequence, and the * that ends it is not searched.
  • The line beside the Query type chips says when a PIR header gave the type, and a chip that says otherwise is refused under the query box. A record without its closing * is refused too, so a FASTA name with a semicolon is never misread.
  • Fixed: a PIR protein record whose description held only letters was searched with those letters before its sequence. Import a database on the local workbench takes PIR files, and a case report prints a PIR query by its name and letters.

Sequences 0.67.0

Search engine 0.19.0

A case keeps an application's claims as filed, as amended, as published and as granted, shows what changed between two sets, and carries your decisions over where a claim's words did not change.

  • New: each claims import in a case is a set of its own, as filed, as amended, as published or as granted, with its date. A publication's XML gives both. A new set becomes current unless it is dated before the current set. See Sets of claims.
  • The Claims step lists each set with its source and SHA-256, and Make current makes another set current, with an optional note for the log.
  • What changed compares any two sets claim by claim and marks each claim unchanged, amended, added or cancelled. Claims are matched by their words, so 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, the rules find its features again. On a claim whose words did not change, even one renumbered, each feature keeps its decision, note and edit, including features the model suggested or you added. The model is not asked again.
  • A locked plan stays locked and follows the set it was locked on. To search the current set, unlock the plan with a reason and draft it again. The report names the set the plan follows, with its label, date, source and SHA-256.
  • New source: Word or text file, up to 15 MiB. In a Word file marked up as an amendment, struck-through text and tracked deletions are left out, and underlined text and tracked insertions are kept. In a claim marked "(Currently amended)", text in [[double brackets]] is left out, in pasted text too.
  • The decisions file names the set of claims behind each import, decision, model call and plan version, and lists each change of the current set with its note.
  • For programs: POST /api/cases/{id}/claims takes file, label, date and current, and GET /api/cases/{id}/claims/sets and GET /api/cases/{id}/claims/changes list the sets and what changed. The OpenAPI description gives every field.

Sequences 0.66.0

Search engine 0.19.0

Admins can export the audit log of sign-ins, API tokens and starts, with every case's log, as JSON lines for a SIEM, and check its hashes with a script from the installation guide.

  • New: with sign-in on, the service keeps its own audit log on the data volume: each sign-in, refused sign-in, failed sign-in and sign-out, each API token made or revoked, each start with its settings, each change to the invite file and each audit export. See Audit export for a SIEM.
  • A secret in the settings shows only as set. The log holds no IP addresses, queries, sequences or results, and nothing rotates or rewrites it.
  • A failed sign-in is noted only when your provider's answer comes back to a sign-in this server started, so a call to /api/auth/callback from anyone else adds nothing. During a burst, the rest of a minute's failed sign-ins are noted as one signin.failed.suppressed event with their count.
  • For admins: GET /api/admin/audit.jsonl?day=YYYY-MM-DD, or with from and to times, gives the audit log and every case's log for a UTC day or a time range as JSON lines. The container's audit-export command writes the same lines from the data volume, which it only reads.
  • Case events keep their IDs, counts and hashes and name each field they leave out, since claims pasted into a case can be from an application that is not yet published.
  • Each event carries the SHA-256 of the event before it and its own, and each log's lines end with the service's check of its whole chain. The installation guide's check_audit.py checks exports with Python 3 alone, across days, and against the data volume.
  • For programs: the OpenAPI description lists GET /api/admin/audit.jsonl.

Sequences 0.65.2

Search engine 0.19.0

A fix to result-file imports: a FASTA36 36.3.8i file with comment lines imports into a case.

  • Fixed: a FASTA36 tabular file in -m 8 with its comment lines, from FASTA36 36.3.8i, imports into a case. Its comment lines name 14 columns, query length and subject length among them, but each row holds the usual 12, and Sequences refused the file at its first row. Sequences now reads the 12, and the import's warnings say so. See Result files from other programs.
  • On the command line, blastn runs megablast unless you add -task blastn, and megablast can miss weaker matches. The guide now says so: add -task blastn when a file should compare like for like with Sequences' own search.

Sequences 0.65.1

Search engine 0.19.0

Cards, notes and selected rows show their state with a soft fill on rounded corners, not a colored edge down one side, and their words still say it.

  • Blocks that show a state now show it with a soft fill on rounded corners instead of a colored edge down one side: a claim feature in Review, the lock banner in Plan, a sentence in Why each reference counts, the broken event in Log, a CDR group of an antibody search and a note in these guides. Their words still say the state.
  • In the results list, a selected row and its pane no longer have a blue edge down their side: the row's tint and bold name show it. A chosen match in a pane fills with a stronger blue. The claim's words in Claims and References sit in a rounded box.
  • The clinical context and funding parts open with their lines set under the part's name, with no rule down their side.

Sequences 0.65.0

Search engine 0.19.0

A case gives short summaries of its searches, its top references and each claimed SEQ ID NO, filled in from fixed templates and the case's own facts, and the report prints the ones you accept or edit.

  • New: summaries in a case. Claims gives each claimed SEQ ID NO's summary: the claims that recite it, and the paragraphs of the description that cite it, quoted. References gives a summary in each open row of the first 5 references and the ones you keep, and one for each search after the features. See Summaries.
  • No language model writes them. Each is filled in from a fixed template and the case's own facts each time you open it, so the same facts always give the same words. Each sentence names its source, and every number, date, SEQ ID NO and publication number in a summary is checked against those facts. A search's summary describes what the search returned, before your marks.
  • Choose Accept, Edit or Reject for each summary. An edit is one paragraph in your own words, and the draft stays beside them. Each summary names the template that filled it in, and 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, such as a new prior art date, the summary asks you to decide again and leaves the report until you do. The case log records every decision, and the decisions file lists each one with its template.
  • The report prints the summaries you accepted or edited in a new Summaries section after By applicant, each with the template that filled it in and who accepted it, and when.
  • A reference in a publication that no record names now reads "Publication unknown", with its sequence's accession, in References and its CSV.
  • For programs: GET /api/cases/{id}/summaries gives a case's summaries, and POST /api/cases/{id}/summaries/decisions accepts, edits or rejects one. The OpenAPI description lists their fields.

Sequences 0.64.1

Search engine 0.19.0

A security update to the container image's runtime.

  • For IT staff: the container image's runtime is updated to a Node.js 22 security release. Nothing changes for people who search, and the install needs no new setting. See What you need.

Sequences 0.64.0

Search engine 0.19.0

A case's decisions download as CSV and JSON, each tied to its event in the case log, and admins can measure the claim rules and model prompts across cases.

  • New: Download decisions (CSV) and Download decisions (JSON) in a case's Report step save every decision in the case: each claim feature, each accept, reject and edit with who made it and when, each model call with the suggestions kept or dropped and why, each sentence decision and the plan's versions. See The decisions file.
  • Each decision gives the number and SHA-256 of its event in the case log, so you can check the file against the log, and the JSON file gives the result of the log's check. An edit gives the finding before and after it. Saving the files changes nothing in the case.
  • For admins: GET /api/admin/decisions.csv and GET /api/admin/decisions.json give the decisions of every case by rule, kind, source, status, prompt and drop reason, with totals. They hold no claim words, quotes, notes or edits, since claims pasted into a case can be from an application that is not yet published. See Usage.
  • For programs: GET /api/cases/{id}/decisions.json and decisions.csv give a case's decisions, and the OpenAPI description lists their fields and columns.

Sequences 0.63.0

Search engine 0.19.0

A case takes the result file of a search you ran outside Sequences, from BLAST+, FASTA36 or another program, and joins its matches to the stored patent sequences.

  • New: bring the result file of a search you ran outside Sequences into a case, under Import a result file in the Report step. Sequences joins each match to the stored patent sequences, and the file opens like one of your own searches, with marks, families, dates and explanations. See Result files from other programs.
  • It reads BLAST+ tabular, XML and JSON output, FASTA36 tabular output, and CSV files with a header row, the CSV files Sequences saves among them. A file can be up to 200 MB and 1,000,000 rows, and the page compresses it before sending it.
  • A match joins by its accession, or by publication number and SEQ ID NO however the number is written. A row that does not join is never dropped: Rows that did not join lists each with its reason, and Download them as CSV saves them.
  • A search of a copy of the patent sets on your own computer joins by each sequence's place in that copy, when the file names the copy's folder. BLAST+ XML, JSON and tabular output with comment lines name it, and so does FASTA36 tabular output with comment lines.
  • References combines an import's matches with your other searches by publication and SEQ ID NO, and the report names each imported file with its SHA-256, how many rows joined and the first rows that did not, each with its reason.
  • For programs: POST /api/cases/{id}/imports takes a result file as the request body, and GET /api/cases/{id}/imports/{jobId}/unresolved, or unresolved.csv, lists the rows that did not join. GET /api/status gives the limits and formats under imports.

Sequences 0.62.1

Search engine 0.19.0

Two fixes to cases: a fragment length written in words reads as one number, and a model's entry in a case's Log keeps time order.

  • Fixed: in Review, a fragment length written in words, such as "at least twenty-five consecutive residues", or with a comma, such as "at least 1,000 contiguous nucleotides", reads as one number, as the claim writes it. See Cases.
  • Fixed: "up to", "at most" and "no more than" give a longest length with no shortest one. "Up to 100 contiguous amino acids" reads as written, and the plan gives "parts of up to 100 amino acids" as the search's reason.
  • In References, a claim's fragment length is the shortest alignment that counts for Meets, so "at least twenty-five" needs an alignment of 25. A length with no shortest one, such as "up to 100", sets none, so identity counts over the whole query. See The References step.
  • Cases you already have keep their findings, plans and References as they were. To correct a length in an older case, choose Unlock the plan, then edit the finding in Review. Claims you import from now on are read the new way.
  • Where the server offers model suggestions, a model's entry in a case's Log now shows the time the model answered, so the Log stays in time order. Model suggestions stay off unless the server's operator names a model.

Sequences 0.62.0

Search engine 0.19.0

On a server you run, the search page and the admins' usage page name the license by its plan and last day, such as "Evaluation license, ends 2027-02-28".

  • On a server you run, the search page's footer names the license by its plan and last day, such as "Evaluation license, ends 2027-02-28". It reads "starts" before the first day and "ended" after the last. See The license file.
  • For admins: the usage page opens with the same line. From 60 days before the end it adds the license's message, in amber, and once new searches stop it turns red. See Usage.
  • The installation guide says what an evaluation license covers and how it ends: read-only, as after a trial, so results stay open and new searches need a new license file.
  • For programs: the license in GET /api/status, and in GET /api/admin/usage for admins, has a label with the plan and the day the license ends.

Sequences 0.61.0

Search engine 0.19.0

Paste a query or drop a file, and the search page reads what it is and where it goes.

  • New: Query type reads DNA / RNA or Protein from what you paste, and Search in follows it: the query's own sets first, then the other type's under Translated search. A click on a chip sets the type yourself. See Query type.
  • The search-type buttons are gone. Antibody CDRs is now a switch beside the search card's title.
  • A fault in the query shows under the box, names the line and the character, and selects it when you run the search.
  • Drop a file on the search card, or choose Open a file: FASTA fills the query box, a sequence listing opens its SEQ ID NO picker, and a publication's XML with claims starts a case with its claims and its listing. See Drop a file.
  • A finished search opens its result by itself, unless you opened another result or are typing in a field. The tab's title says Running or Done until then, and after a reload the status bar follows your newest search. See While a search runs.
  • Each result has an address that opens it again for your account. Copy link copies it, and Back closes the result. See Addresses.
  • Recent jobs has Find a search, by name or set, and a status choice.
  • A result with no match offers next steps, each a button that fills the search form. See No match.

Sequences 0.60.0

Search engine 0.19.0

Review can ask a language model your server runs for claim features the rules missed, with the details a search needs and the paragraphs of the description that bear on the claims.

  • New: where the server's operator names a language model, Review can ask it for claim features the rules missed. The button reads "(Pilot)", and each suggestion counts only once you accept it. Suggestions stay off until a model is named. See Claim suggestions from a model.
  • A suggestion can carry the details a search needs: SEQ ID NOs, an identity threshold, a range of positions or a fragment length. Each number must be written in the claim's own words, or the details are dropped and the suggestion kept. Accepted with details, a suggestion drafts a search as a rules finding does.
  • When claims come from a publication's XML, the case also reads the description's numbered paragraphs. The model gets up to 40 of them that cite a SEQ ID NO the claims name or define a term the claims use, 60,000 characters at most, and a suggestion may quote the paragraph it draws on.
  • A model on processors can take minutes for one case. The call runs on the server while you keep working, the step shows how long it has been asking, and the notice says what it brought. The server asks the model for one case at a time.
  • Each call is recorded in its case, whether it brings suggestions or fails: the request as sent and its SHA-256, the model's answer, how long it took, and each suggestion kept or dropped with the reason. The case's log names each call.
  • For IT staff: BIOSEQUENCE_CLAIMS_MODEL_API=openai-chat sends a chat completions call, the format vLLM, llama.cpp's server and Ollama take, and the service waits up to 900 seconds for it unless BIOSEQUENCE_CLAIMS_MODEL_SECONDS says otherwise. See Claim suggestions from a model.
  • For programs: POST /api/cases/:id/suggestions with background: true answers at once, and GET /api/cases/:id/suggestions reads the call's state until it ends. Without background the request waits for the answer, as before.

Sequences 0.59.0

Search engine 0.19.0

A case lists every reference its searches found under the claim feature that found it, ranked by fixed rules, in a new References step.

  • New: a case's References step, between Run and Report, lists every reference the case's searches found under the claim feature whose search found it, with the claim's words and its threshold. See The References step.
  • Inside a feature, references rank by fixed rules, with no language model: whether the best match meets the claim's threshold over the whole query, how much of the claimed region it covers, its date against the prior art date, then Query Match and E-value. Each row shows its rank with each score beside it.
  • Open a row for its scores in words, such as "Meets 90%: 97.3% identity over the full query", the search that found its best match and Open results. A feature with many SEQ ID NOs says how many each reference matches, such as "Matches 37 of 250 SEQ ID NOs".
  • Show excluded lists the references that every search that found them excluded, and Download CSV saves every reference of every feature with its scores.
  • For programs: GET /api/cases/:id/references gives the same list, one feature at a time with feature, offset and limit, and GET /api/cases/:id/references.csv the CSV.

Sequences 0.58.0

Search engine 0.19.0

The same service as 0.57.0, with a fix in a tool only ConductScience runs: the build of Patent DNA and Patent proteins goes on when the patent office's sequence listing service fails for a publication.

  • When the patent office's sequence listing service answers a publication with a server error, the build of Patent DNA and Patent proteins goes on without it and counts it. A later update asks for it again and adds its sequences once the service sends them. A publication whose listing is printed in the publication itself keeps it meanwhile.

Sequences 0.57.0

Search engine 0.19.0

One search can cover several sets: check them under Search in, and the result opens on one list of references.

  • New: Search in takes several sets. Check two or more, up to 8, and they run as one combined search, with one status bar and one result. A new search starts with its type's sets checked, apart from TrEMBL and NCBI core_nt, which take minutes. See Search several sets at once.
  • For a DNA or protein query, Translated search lists the sets of the other type, so the same search can read a protein in Patent DNA, in all six frames.
  • A combined search opens on References: each patent or database entry once, with its best match across the sets and a chip for each set that found it. Found in and Kind filter the list, and Keep or Exclude marks a reference in every search that found it.
  • Choose a reference to see its matches by set and query, each one click from its alignment. Search record lists each set's search with the SHA-256 of its record, and opens that search by itself.
  • For programs: GET /api/jobs/:id/hits/:row gives one hit of a search by its row number, with its mark. See Programs and API tokens.

Sequences 0.56.0

Search engine 0.19.0

The listing import reads the sequence listing inside a patent publication's full-text XML, and programs can find jobs by name, set or state.

  • New: Import in the Listings tab of Sequence tools takes a patent publication's full-text XML when its listing is inside it, as in publications of 2001 to 2004, and reads that listing. A publication whose listing was filed separately says so. See Listings and the ST.26 check.
  • If an entry of such a listing cannot be read, it is left out and the listing's facts name it, such as "SEQ ID NO: 3 could not be read". The other entries import and search as usual.
  • For programs: GET /api/jobs takes text, to find jobs by name or by the set or database they searched, and status, one or more states separated by commas. GET /api/jobs/:id/request gives the request a search was started with. See Programs and API tokens.

Sequences 0.55.1

Search engine 0.19.0

A fix to sample results: a sample added, removed or replaced while the service runs now shows at once on every server.

  • Fixed: on a Linux server, a sample result added, removed or replaced a moment after another change could keep showing as it was until the next change. Each change now shows at once. See A sample result.
  • For IT staff: sample add and sample remove now write a small .generation file in the samples folder after each change. Leave it there; the service reads it to see the change. See Sample results.

Sequences 0.55.0

Search engine 0.19.0

Sample results: real searches of public sequences, saved and shown read-only, that anyone can read before signing in, linked from the sign-in page.

  • New: sample results you can read before signing in. Each is a real search of a public sequence, such as GLP-1 against Patent proteins, saved and shown read-only, with its patents grouped by family and each row opening to its best alignment. See A sample result.
  • The sign-in page says "Not ready to sign in? See a sample result." when a server has one. A sample takes the Patents tab's choices: families or every publication, its sorts, and Published before, which folds the later publications and counts them.
  • For IT staff: sample add makes a sample from a finished search of a patent set, and sample list and sample remove manage them. Anyone who can reach the server can read a sample, so use searches of public sequences. See Sample results.

Sequences 0.54.0

Search engine 0.19.0

The search page now offers suggested searches: real public sequences, such as GLP-1 and human insulin's gene, that run with one click and show what to try in their results.

  • New on the search page: Suggested searches, real public sequences you can search with one click. GLP-1 against Patent proteins, human insulin's gene against Patent DNA, green fluorescent protein against Patent DNA read as protein, the antibody trastuzumab by its CDRs, and the peptide SIINFEKL as exact text. See Suggested searches.
  • Run fills the form as you would and starts the search, and its result opens when it finishes. Edit first fills the form and leaves the search to you. Each card links the public record its sequence comes from.
  • The result of a suggested search shows Things to try: two or three steps that use it, such as setting a prior art date. Close hides them for that search in this browser.
  • 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.

Sequences 0.53.0

Search engine 0.19.0

Patents now groups publications by patent family, marks a family as one, and folds the publications after your date into one line under the list.

  • On Patents, the new Group choice sets what a row is. With Families, each row is one patent family: the publications that share a priority claim or an application, or the same title and applicant where no priority claim is on record. The row shows the earliest publication, the family's best measures and a chip such as "+2 in family". See By patent.
  • The line over the list counts both, such as "12 publications in 5 families match these filters." Every publication lists one publication a row, as before. Applicants and a case's report count publications whatever Group says, and Export matches saves one row for each publication.
  • A family's pane says how the family was formed, lists its publications earliest first with their dates and an Open link, and shows the chosen publication's matching sequences with the best alignment.
  • Keep and Exclude on a family row mark each of its publications in the list at once. The Mark column says "Kept" when the whole family is kept and "Some kept" when only part of it is. To mark one publication of a family, choose Every publication. See Keep and exclude.
  • With Published before set, Patents keeps the publications from before your date, the US and WO publications filed or claiming priority before it, marked "Filed before", and the undated ones. The later publications of the same sequences fold into one line under the list, and Show them brings them back, marked "After date". See Dates.
  • A Smith-Waterman or global alignment search that reaches its limit now says on the page that its list holds the best-scoring sequences, and that Exact text finds every exact occurrence.

Sequences 0.52.0

Search engine 0.19.0

Programs can now run one search across several sets and get one list of what they found, where a patent or database entry that several sets found is one row.

  • In the API, POST /api/groups runs one search across 2 to 8 sets, such as Patent proteins, Patent DNA and Swiss-Prot. Every set is checked first: if one cannot take the query, or the queue lacks room, none starts and the answer says why. See Programs and API tokens.
  • GET /api/groups/{id}/references gives one list for a combined search. A patent or database entry that several sets found is one row, with its best measures, its earliest date, the sets and queries that found it, and each match. It takes the filters, sorts, families and date fold of a search's patents, plus set and kind.
  • The list answers while some sets are still running and says which searches it holds. GET /api/groups/{id} gives each search's state, matches and time left.
  • POST /api/groups/{id}/marks keeps, excludes or clears up to 500 references at once, in every search of the combined search that found them, so each search's own lists show the mark.
  • GET /api/groups/{id}/references.csv has one line per reference, with a sets column and the family columns. POST /api/groups/{id}/again runs a combined search again, all of it or only the sets you name, and DELETE /api/groups/{id} stops its searches still waiting or running.
  • Only the account that made a combined search can see or change it. Combined searches are in the API only for now; the page shows them in a later release.

Sequences 0.51.0

Search engine 0.19.0

Search results are now a list of one-line rows beside a detail pane, so you can read a result, mark it and move to the next without leaving the list.

  • On Patents and Alignments, each result 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, and on a narrower screen the pane opens under the row. The row you select stays where it is on the screen. See Results.
  • Keys work while the list has the focus: Up and Down move, Page Up, Page Down, Home and End jump, K keeps, X excludes, U clears, R goes to the reason field, Esc comes back and C copies the alignment. Keys, above the list, lists them.
  • The pane shows the alignment 60 letters to a line, with each letter of the hit that differs from the query in bold. Copy alignment and Download .txt give it as text.
  • A patent's pane lists its matching sequences by SEQ ID NO and opens on the best one's alignment. Choose another sequence to see its alignment.
  • An alignment's pane lists where its sequence is published, earliest first, and links each database entry that holds it to its record at UniProt or ENA.
  • Keep, Exclude and Clear are in the pane, beside Reason (optional). To give a marked row a new reason, type it and press Enter. See Keep and exclude.
  • In the API, GET /api/jobs/{id}/hits takes publication for the hits of one publication, and GET /api/jobs/{id}/publications takes row for every publication of one row's sequence. See Programs.

Sequences 0.50.0

Search engine 0.19.0

A case now keeps its application number and prior art date, a plan can search every set on the server, and a plan of any size runs to the end by itself.

  • A case keeps its Application number and Prior art date. A publication's XML fills them in, and when it gives an earliest priority date you can choose that date instead. Change in the case's header edits both. See Cases.
  • New plan searches look for what was published before the case's prior art date, and the date sentence under Why each reference counts names the date it uses. A locked plan version keeps the date it was locked with.
  • A drafted plan searches every set on the server that holds the finding's type, the patent sets first: for a protein, Patent proteins and Swiss-Prot. Set to search lists every set, from any snapshot on the server, and each attached database.
  • TrEMBL and attached databases such as core_nt join a plan only when you ask: a button such as Add TrEMBL adds a copy of each fitting search and says how long they should take.
  • Any locked plan runs, however many searches it needs. The case starts what the queue has room for, keeps 2 places for searches from the form, starts the rest as searches finish, and goes on by itself after the server restarts.
  • Each run shows a line such as "7 of 12 searches done, about 4 minutes left." A search that failed, was cancelled or was interrupted has Run this one again.
  • The case report adds the application number and prior art date, a table of the plan's searches and the features no search covered.
  • In the API, PATCH /api/cases/{id} takes applicationNumber, priorArtDate and priorArtDateBasis, and GET /api/cases/{id}/runs gives each run's progress. On a server you run, BIOSEQUENCE_MAX_JOBS sets how many searches the queue holds. See Settings.

Sequences 0.49.0

Search engine 0.19.0

Publications that tie now list the earliest first, the date filter keeps a sequence whose US or WO publication was filed before your date, and programs can list a search's patents by family.

  • On Patents, publications that tie on the sort now go earliest first, undated last, so among the publications with the best Query Match the earliest is on top. Under the earliest or latest date, publications of one date go highest Query Match first. See Filters and sorting.
  • On Alignments, rows that tie go by their sequence's first published date, oldest first. Before this release, ties on both tabs went by publication number or row number.
  • Published before now also keeps a sequence first published on or after your date when one of its US or WO publications was filed, or claims priority, before it, where the record gives those dates. Its publications from on or after your date are still marked "Published on or after your date". See Dates.
  • On Patents, Export matches (CSV) ends with four new columns: family, family_size, family_basis and date_group. See Exports and search records.
  • In the API, GET /api/jobs/{id}/publications takes group=family for one row per patent family and later=hide to fold the publications from on or after the date. Every patent row gives dateGroup, and a hit row kept by a filing or priority date carries filedBefore. See Programs and API tokens.
  • POST /api/jobs/{id}/marks and a case's marks route take ids, up to 500 publications at once, so a program can keep or exclude a whole family.
  • Family rows and the date fold are in the API only. The page still lists one row per publication.

Sequences 0.48.0

Search engine 0.19.0

The funding trail can show what a hit's US published applications state about federally sponsored research, word for word, after its papers, on a server whose settings name the add-on and whose funding trail folder holds the statements.

  • On the Alignments tab, a muted line after a hit's papers names the statements on federally sponsored research of the US published applications that hold its sequence, earliest first. Under it, a closed part for each of up to three applications has the application's number and publication date for its summary. See Funding trail.
  • Opened, a part shows the statement word for word, under the heading the application puts above it, if there is one. A hit whose applications give more ends on a line that counts the rest. A statement that only says it does not apply is left out. The parts match no statement to an NIH award or to a paper's grants, and draw no conclusion.
  • An add-on runs only on a server whose settings name it, and a server that names none holds none of its data.
  • On a server you run, the funding trail folder is now format 1.1.0 and can hold the statements, and each week a new release of the folder adds that week's applications. A folder of 1.0.0 still opens and shows none. Upgrade the image before you load a folder with statements: an older image loads it and shows none of them. See Funding trail.
  • In the API, where the folder holds statements, a hit row carries statements, at most three, each with publication, published, heading and paragraphs, and statementsMore when its applications give more. See Programs and API tokens.

Sequences 0.47.1

Search engine 0.19.0

The same service as 0.47.0, with a fix in a tool only ConductScience runs. The hosted service took 0.47.0's changes with this release.

  • Nothing changed for people who search, and a server that runs 0.47.0 does not need this release. The 0.47.0 notes list what both releases add.

Sequences 0.47.0

Search engine 0.19.0

On Alignments, a patent hit names the applicant of its earliest publication. The CSVs gain that publication and a patent publication's filing date, application number and priority claims, and the case report a Filed column, where the records give them.

  • On Alignments, a line under a patent hit's title names the applicant of the publication the title names, the sequence's earliest. Export matches (CSV) ends with that publication: first_publication, first_kind, first_applicant, first_filed, first_application_number and first_priority. See Exports and search records.
  • On Patents, the CSV ends with application_number, filed and priority. A priority cell lists every claim a record gives, earliest first, each as a date and a number.
  • Where a publication's record gives its filing date or priority claims, its row on Patents shows them beside the publication date, and the case report adds a Filed column. They do not change its date. The records in the patent sets give neither. See Dates.
  • In the API, a hit row has firstPublication and a publication row has applicationNumber, filed and priority. See Programs and API tokens.

Sequences 0.46.0

Search engine 0.19.0

The funding trail, a second add-on: under each paper of a hit, the grants PubMed lists for it, with the NIH awards they match, on a server whose settings name the add-on.

  • On the Alignments tab, a paper under a hit whose grants PubMed lists has a closed Funding part. Opened, it shows each NIH award the paper's grants match: the project number linked to NIH RePORTER, the title, each institution with its fiscal years and the investigators NIH lists. See Funding trail.
  • The paper's other grants follow as PubMed lists them: other funders, contracts and numbers that match no single NIH project. The part names its sources and their dates, shows no award amounts and draws no conclusion. Sources and terms has a card for them.
  • An add-on runs only on a server whose settings name it, and a server that names none holds none of its data.
  • On a server you run, BIOSEQUENCE_ADDONS takes funding beside clinical, and BIOSEQUENCE_FUNDING names a funding trail folder from ConductScience, which load checks file by file. See Funding trail.

Sequences 0.45.0

Search engine 0.19.0

Clinical context, the first add-on: under a protein hit that is a drug target, the drugs that act on it, the diseases they are studied for and their clinical trials, on a server whose settings name the add-on.

  • On the Alignments tab, a Swiss-Prot or TrEMBL hit whose UniProt entry is a drug target has a closed Clinical context part under its title. Opened, it lists the drugs that act on the target, highest clinical stage first, with each one's mechanism; the diseases each drug is studied for; and their trials, ten at a time. See Clinical context.
  • The target, the drugs and the diseases link to Open Targets, each drug also to ChEMBL, and each trial to ClinicalTrials.gov. The part names its sources and their dates and draws no conclusion from them. Sources and terms has a card for them.
  • Patent and EST hits, hits with no UniProt accession and antibody region searches show no clinical context.
  • An add-on runs only on a server whose settings name it, and a server that names none holds none of its data.
  • On a server you run, BIOSEQUENCE_ADDONS and BIOSEQUENCE_CLINICAL turn it on with a clinical context folder from ConductScience, which load checks file by file. See Clinical context.

Sequences 0.44.0

Search engine 0.19.0

A user guide, an installation guide and these release notes, served by every server for the release it runs.

  • The pages open without sign-in on the hosted service, on a server you run and on the local workbench, and the docs home names the release the server runs.
  • The User guide covers searches, results by patent, applicant and alignment, papers, dates, exports, watches, portfolio exports, cases, listings, antibody CDR search and API tokens.
  • The Installation guide is for the IT staff who run Sequences on your own server, or in your AWS account with the AWS install kit.
  • On a server with its own sign-in, watches now send email to people who sign in with a verified email address, as the installation guide says. A watch saved before this release sends none until you watch its search again. See Watch emails.
  • Every page links to the docs from its footer.

Sequences 0.43.1

Search engine 0.19.0

A search that found no patent publication, such as a search of Swiss-Prot, now opens on its alignments instead of on an empty Patents tab.

  • A search opens on Patents only when it found a patent publication, and otherwise on Alignments, with the Patents tab beside it. Searches run before this release open the same way. See Results.
  • Nothing else changed.

Sequences 0.43.0

Search engine 0.19.0

The AWS install kit: one CloudFormation stack puts Sequences on a server in your organization's own AWS account. A server with sign-in also answers a load balancer's health check at its IP address, and every server keeps an idle connection open longer than an AWS load balancer does.

  • A CloudFormation template and kit.sh, the script its server runs, put Sequences on one server in your organization's own AWS account (kit 1.0.0). See In your AWS account.
  • One stack creates a memory-optimized EC2 server with no public address and no key pair, encrypted gp3 disks, its security groups and its role, and, with sign-in, an HTTPS Application Load Balancer with your ACM certificate. The role can read only its folder of your bucket, the stack's settings and the secrets you name, and with ECR pull from one repository.
  • On its first boot the server installs Docker, gets the Sequences image from S3 or ECR, copies each snapshot from S3, loads it against its SHA-256 list and runs verify on it before the service starts. The stack completes only then.
  • You manage the server through Session Manager: kit.sh apply puts a stack update in place, and kit.sh check checks the install. ServerPolicyArns gives the server's role your own policies from its first boot, PatentLink sets or removes the link on each patent number, and AddOns takes only none: kit 1.0.0 runs no add-ons.
  • The kit runs 0.43.0 and later.
  • With sign-in on, GET /api/health also answers at the server's IP address, because a load balancer's health check cannot name the host. It still says only whether the service can search. Any other path, or a host name that is not allowed, still gets 403. See Health and logs.
  • Every server keeps an idle connection open for 65 seconds, longer than an AWS load balancer's 60.
  • Nothing changed for people who search.

Sequences 0.42.0

Search engine 0.19.0

Applicant watches and portfolio exports: a watch can follow an applicant as well as a search, one applicant's publications export with their sequences, and the case report lists the kept results by applicant.

  • Each named row of a snapshot search's Applicants tab has Watch this applicant, and Your watches has a Watch an applicant form that takes a name. The watch follows the applicant on its snapshot, with the names your account has joined to it. A watched row reads Watching, which goes to the watch. See Watches.
  • After each weekly addition, a watch on an applicant lists the applicant's new publications under Your watches, each with its date and number of sequences, linked to the patent. It runs no search. On a server with watch emails set up, the new publications also go to your email. The hosted service sends no watch emails.
  • Your watches now shows whenever the server has a patent snapshot, so the form is there before your first watch.
  • Export portfolio, on an Applicants row or on a watch, starts a job in Recent jobs. It collects every publication in the snapshot and its weekly additions whose first-named applicant is the applicant or a name you joined to it, with each of their sequences in Patent DNA and Patent proteins, as a CSV file and a FASTA file. See Portfolio export.
  • An export holds up to 100,000 sequences and 100 million letters, oldest publications first, and says how much of the applicant it holds. Its files download from the row or from the job's card.
  • The PDF and Word reports of a case have a new By applicant section after the table of searches. It groups the publications you kept, and the publications that hold each kept match, by applicant across all of the case's searches, with your joins.
  • For programs, POST /api/watches also takes an applicant, and POST /api/portfolios, GET /api/jobs/{id}/portfolio.csv and GET /api/jobs/{id}/portfolio.fasta are new. All are in the OpenAPI description.
  • On a server you run, a portfolio export counts toward BIOSEQUENCE_CONCURRENCY, stops at BIOSEQUENCE_JOB_SECONDS and keeps its files with its job, about 250 MB at its limit. It needs no new setting. After a license or trial ends, no watch or export starts. Finished exports still download, and watches can still be paused or deleted. See Settings.

Sequences 0.41.0

Search engine 0.19.0

Why each reference counts: a case drafts short sentences on why each publication it keeps bears on the claims, from facts it has stored, and you accept, edit or reject each one.

  • For each publication you keep, and for the top 5 publications of each search, the case drafts up to four sentences: the best alignment, the claim's identity threshold that the search used, the publication date against the filing date, and the genes and diseases named by the paper that reports the sequence. See Cases.
  • Each sentence comes from a fixed template and names its source, with a link to it: the result row, the plan version, or the publication's or paper's page. No language model writes them, so the same facts always give the same words.
  • Under Why each reference counts in the Report step, accept, edit or reject each sentence. The PDF and Word reports print the accepted and edited sentences of the publications you keep, with their sources. The case log records each decision, and an edit keeps both the draft and your words.
  • When the facts behind a sentence change after you decide, the sentence says that its draft changed, and it leaves the report until you decide again.
  • Once a license or free trial has ended, a new decision is refused. The drafts and decisions can still be read, and reports still made.
  • Two fixes to the PDF and Word reports. Since 0.32.0, a case whose claims came from a publication's XML file printed "[object Object]" in the Claims row, and the row now names the publication. Where an event in the report's case log quotes a SHA-256, the report now prints its first 12 characters, so the line no longer wraps. The case keeps the full value.
  • For programs, GET /api/cases/{id}/explanations lists the drafts and decisions, and POST /api/cases/{id}/explanations records a decision. Both are in the OpenAPI description.
  • On a server you run, no setting and no stored data change, and existing cases get drafts from what they already hold. The paper sentence needs a paper folder made with PubTator 3, which the papers command of 0.40.0 already takes. A folder made without it must be made again, with a new ID for BIOSEQUENCE_PAPERS. See Paper citations and marks.

Sequences 0.40.1

Search engine 0.19.0

The same service as 0.40.0, with two fixes. The hosted service took paper citations, paper dates and retraction marks with this release.

  • Nothing changed for people who search, and a server that runs 0.40.0 does not need this release. The 0.40.0 notes list what both releases add.

Sequences 0.40.0

Search engine 0.19.0

Each paper on a hit reads as its PubMed citation, is dated by the day it first came out, and says whether it was retracted, drew an expression of concern or was reinstated.

  • The hosted service never ran this release. It took these changes with 0.40.1.
  • Each paper under a hit reads as a full citation, with links to its PubMed record and its DOI. See Papers.
  • A paper is dated by the day PubMed says it first came out, online or in print, whichever came first. Published before compares days, so a hit whose only paper came out in December 2015 no longer passes a date of June 1, 2015. See Dates.
  • A paper that was retracted, drew an expression of concern or was reinstated says so 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 mark says so. A paper whose PubMed ID was deleted keeps its year and says it is not in PubMed.
  • The CSV of the alignments adds paper_dates and paper_marks. The case report lists the papers of the matches you kept in the same way, and names the PubMed data each search used.
  • A search keeps the papers it had when it ran, so a search from before this release shows no marks. Run it again to see them.
  • For programs, each hit from GET /api/jobs/{id}/hits carries the same papers in paperList, with paperDates and paperMarks as in the CSV.
  • On a server you run, BIOSEQUENCE_PAPERS names a paper folder made with the new papers command from PubMed's files and the Retraction Watch Database. Without it, each paper keeps its year. See Paper citations and marks.

Sequences 0.39.0

Search engine 0.19.0

Results by applicant: a new Applicants tab groups a search's patents by who applied, lets you join names that differ, and suggests the matching record in the Research Organization Registry (ROR).

  • A search of the snapshot has an Applicants tab beside Patents and Alignments. Each row is one applicant, the first applicant each publication names, with its publications, earliest date and best identity. See By applicant.
  • Spellings that differ only in case, accents, punctuation or a legal ending such as Inc or GmbH share a row. Publications with no applicant are listed under "Applicant not given".
  • Names that still differ, such as a company's old and new names, can be joined: select them and choose Join selected names. A join covers all of your searches until you undo it. Every join and undo goes into your account's join log, and Check the log verifies it.
  • Where an applicant's name matches an organization's name in ROR, the row links to its ROR record. A suggestion never joins or renames names. The hosted service suggests from ROR v2.15, data of October 20, 2026.
  • The filters work as on Patents, and Export matches (CSV) saves the rows as the screen shows them.
  • For programs, GET /api/jobs/{id}/applicants and GET /api/jobs/{id}/applicants.csv are in the OpenAPI description. Joins are made on the page only.
  • When a sequence listing gives its applicant's name in another script, such as Japanese or Korean, and also in Latin letters, new snapshots store the Latin form.
  • On a server you run, BIOSEQUENCE_ROR names a ROR name index made with the new ror command. Without it, the tab works with no suggestions. Joins are kept in applicants/ in the data volume, so back it up. See ROR suggestions.

Sequences 0.38.0

Search engine 0.19.0

Programs can use Sequences: personal API tokens, batches of up to 100 searches, and an OpenAPI 3.1 description of the routes.

  • Signed in with a browser, open API tokens beside your account to make a named token, which the page shows once. A program sends it as Authorization: Bearer <token> and acts as your account, under the same rules.
  • A token lasts at most 90 days, and an account may have 10 live tokens. The page shows when each token was last used and lets you revoke it.
  • 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.
  • GET /api/openapi.json describes the routes a program uses. It is open without sign-in and holds no data.
  • On a server you run, BIOSEQUENCE_TOKEN_DAYS sets the longest a token may last, and 0 turns tokens off. Admins see every account's tokens on the usage page and can revoke any of them. See API tokens for programs.

Sequences 0.37.0

Search engine 0.19.0

Hits show their papers and database entries, with dates, and two large public sets join the snapshot: UniProtKB/TrEMBL and the ESTs of the European Nucleotide Archive (ENA).

  • Under each hit's title the page says how many database entries and papers hold the sequence, and lists up to three papers, oldest first, each linked to PubMed or its DOI.
  • Database entries now keep the day their sequence became public, where it is known, so Published before uses a hit's earliest date across its patents, entries and papers. A paper is dated by its year only and counts as earlier than every day of that year, so the filter never hides a paper from the year of your date. See Dates.
  • TrEMBL and the ESTs are new sets. The ESTs come from ENA's files of September 10, 2026. Swiss-Prot entries now list their papers.
  • The hosted service also searches NCBI core_nt with BLAST. When ENA also holds a hit, the hit shows ENA's description and papers and, if that record is at version 1, its first public day.
  • TrEMBL and core_nt searches take minutes, so the page never selects them for you, and it says how long a search should take before you run it.
  • The CSV of the alignments ends with four new columns: entries, papers, pubmed_ids and dois.
  • A Smith-Waterman or global search of a whole set can return a slightly different list when it runs again, because its E-values come from a random sample of scores. BLAST, exact and pattern searches return the same hits each time.
  • The sets are listed in one order (patent DNA, patent proteins, Swiss-Prot, TrEMBL, ESTs), each from its newest snapshot. An older snapshot leaves the list, but a saved search, Run again or a watch still finds it.

Sequences 0.36.0

Search engine 0.18.0

The hosted service moved to new servers at the same address. Pages, searches, settings and saved searches did not change.

Sequences 0.35.0

Search engine 0.18.0

The Listings tab checks a WIPO ST.26 sequence listing before you file it.

  • The check reads the file against ST.26 version 1.7 in your browser, so the file does not leave your computer.
  • 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 download the problems as CSV.

Sequences 0.34.0

Search engine 0.17.0

A plans page and free trials on the hosted service.

  • The plans page lists the plans of the hosted service and the server edition, with their prices, and opens without sign-in.
  • On the hosted service, an account that was not invited starts a free trial at its first sign-in, on the terms the plans page gives. The search page shows the days left. When a trial ends, its searches stay open and no new search starts.

Sequences 0.33.0

Search engine 0.16.0

Watches: a finished search of the snapshot runs again when the snapshot gets newer records, and shows only the matches it has not found before.

  • Choose Watch this search on a finished search. Your watches lists each watch with its state and its new matches.
  • Where the server sends email, new matches also go to your address, and each email has a link that pauses the watch without signing in.
  • An account may have 50 watches. See Watches.

Sequences 0.32.0

Search engine 0.15.0

Case reports as PDF and Word files.

  • A case's Report step downloads its searches, kept results and notes as a PDF or a Word file.
  • The server keeps no copy of a report. The case log records each report's SHA-256.

Sequences 0.31.0

Search engine 0.14.0

A usage page for the admins of a server with sign-in.

  • Admins see a Usage link beside their account. The page counts searches and other work for each account and day over the last 7, 30 or 90 days.

Sequences 0.30.0

Search engine 0.13.0

A health address for uptime monitors.

  • GET /api/health answers without sign-in: 200 when the server can search and 503 when it cannot. It says nothing else.

Sequences 0.29.2

Search engine 0.12.2

An update of the hosted service no longer drops a page or stops a search.

Sequences 0.29.1

Search engine 0.12.1

The query box takes a pasted sequence with no FASTA header.

Sequences 0.29.0

Search engine 0.12.0

A My account link beside your account on the hosted service, for the settings of your ConductScience account.

Sequences 0.28.0

Search engine 0.11.0

Antibody CDR search on the search page.

  • Paste up to 10 heavy or light chains and number them by Kabat, Chothia or IMGT, or type the CDRs in, then search the CDRs against a protein set.
  • Results group by patent or by stored sequence and show, for each CDR, an exact match, its identity or that it is missing. See Antibody CDR search.

Sequences 0.27.0

Search engine 0.10.0

The server edition: Sequences as a container image that your IT staff run on your own server.

  • People sign in through your own OpenID Connect provider, such as Okta, Microsoft Entra ID or Keycloak.
  • A license file signed by ConductScience is checked inside the container. There is no license server.
  • Snapshot copies load with a SHA-256 check of every file, and each snapshot carries its sources and terms.
  • The service turns off the usage reports that BLAST+ can send to NCBI. See the Installation guide.

Sequences 0.26.0

Search engine 0.9.0

Claims to search plan, and antibody numbering in the service.

  • A case imports claims and marks what each claim recites: SEQ ID NOs, identity thresholds, fragments, complements, CDRs and sequences written out in the claim. You review each finding, and the reviewed list locks as a search plan before any search runs.
  • The service numbers antibody chains by Kabat, Chothia or IMGT. The search page uses it from 0.28.0.

Sequences 0.25.0

Search engine 0.8.0

Sequence listings, sequence tools, additions to the snapshot, and Run again.

  • The Listings tab reads a WIPO ST.26 or ST.25 sequence listing and searches its sequences.
  • The Editor, Fragments and Align tabs check, edit, save and align sequences.
  • A snapshot can take additions of newer records, and the Main plus New collection searches them with it.
  • Run again repeats a finished search with the same queries, settings and data.

Sequences 0.24.0

Search engine 0.7.0

A new design, and the search page took the name Sequences.

  • The page serves its own font and loads nothing from another site.

Sequences 0.23.1

Search engine 0.7.0

The hosted service moved to sequences.conductscience.org. The earlier address forwards there.

Sequences 0.23.0

Search engine 0.7.0

The hosted service, for invited accounts.

  • Sign in with a ConductScience account. Each search belongs to the account that started it, and no other account can open it.
  • The hosted service searches the snapshot. Importing a collection of your own is for the local workbench.

Sequences 0.22.0

Search engine 0.7.0

A fixed benchmark of searches checks each release, and the search record says that the E-values of Smith-Waterman and global alignment come from a random sample.

Sequences 0.21.0

Search engine 0.6.0

Smith-Waterman and global alignment, and an automatic choice of method.

  • A snapshot search picks its method by the query, and the search record states the pick: Smith-Waterman for a protein, and BLAST with a full exact scan for DNA under 50 bases.
  • You can also choose Smith-Waterman or global alignment by name.
  • Align exactly aligns up to 1,000 chosen rows of a BLAST search again in full, by Smith-Waterman or global alignment, and adds the new scores to those rows and the CSV.

Sequences 0.20.0

Search engine 0.5.0

Exact and pattern search. Both read every stored sequence, so their counts are complete.

  • Patterns take wildcards, gaps, repeats and IUPAC codes, and a DNA search covers both strands unless you pick one.
  • Each row gives the strand, how many times the text occurs and its first 20 positions.

Sequences 0.19.0

Search engine 0.4.0

Results by patent and by date.

  • A snapshot search opens on its patents, each with its office, number, date, applicant and SEQ ID NOs.
  • Published before keeps a sequence first published before your date, and an undated sequence always stays.
  • Keep or exclude a patent or a row, with a reason. The CSV of the patents and the CSV of the alignments both carry the marks.
  • Past searches come in pages, and each search shows an estimate of its time.

Sequences 0.18.0

Search engine 0.3.0

The sequence snapshot: GenBank's patent sequences, the patent proteins of EMBL-EBI and UniProtKB/Swiss-Prot in one read-only collection, each sequence stored once with every record and publication that holds it.

Sequences 0.17.0

Search engine 0.2.0

Searches of large databases, such as the patent sequence sets.

  • Short queries, DNA under 50 bases and proteins of 30 residues or fewer, use BLAST's short-query settings automatically.
  • Each query says what it ran with and whether it reached the limit of 5,000 sequences.

Sequences 0.16.0

Search engine 0.1.0

The first release: a local workbench that runs NCBI BLAST+ on your computer in all four directions between DNA and protein.

  • Import nucleotide or protein FASTA collections, run searches as jobs you can cancel, filter by identity, coverage and text, and export CSV and a full JSON record.