Skip to content

Changelog

Releases of the GeneWeaver web application, newest first.

The version you are using is shown at the bottom of every page on geneweaver.org as Application Version.

Version numbering

A version containing a letter — 1.6.0a, 1.6.1a — is a pre-release that goes no further than our internal test environment. Only plain versions such as 1.6.0 and 1.6.1 reach geneweaver.org. Pre-releases are folded into the entry for the release that carries them, so nothing is described twice.

1.6.2 — 17 September 2026

Fixed

  • Recently created gene sets can be found through the API again. GeneWeaver's programmatic gene set search (/api/genesets/search, used by the newer API and the tools built on it) was answering from an index that had not been rebuilt since May. Gene sets created after that — 25 of them, including every gene set uploaded on 14 September — came back as no results, with nothing to indicate the gene set existed at all. The search now has a rebuild scheduled every night, so a newly created gene set becomes searchable within a day, and the backlog has been cleared.

    Searching on geneweaver.org itself was not affected — the website's own search uses a different index, which was already being rebuilt nightly. Fetching a gene set directly by its ID was not affected either. Only the programmatic search was, and only for gene sets newer than the last rebuild.

  • "Find other GeneSets from this publication" works on large publications. For a publication with thousands of gene sets attached the page never finished loading and eventually returned an error — the GeneWeaver record for the Gene Ontology Consortium paper has 14,799 gene sets, and the page asked the database a separate question about every one of them, twice. It now asks once and shows 100 gene sets at a time, with Previous / Next links and a count of how many there are in total. The page went from not loading at all to well under a second.

  • A gene set with no publication no longer returns an error. More than half of all gene sets have no publication attached, and for those this page failed outright rather than saying there was nothing to show. It now shows the same "no other GeneSets" message as a publication with a single gene set.
  • The list now shows only the gene sets you are allowed to see. Permission was being checked and then ignored, so on publications that mix public and restricted gene sets the page listed the name, description, size, tier and species of sets the signed-in viewer had no access to. Gene-level data was never exposed, and the page has always refused to render for visitors who are not signed in. The count now reflects what you can see, and asking for a gene set you cannot read no longer reveals which publication it belongs to.

The three Find other GeneSets from this publication fixes are long-standing — that page has behaved this way since it was added in 2015 — and are unrelated to the 1.6.0 and 1.6.1 releases.

1.6.1 — 16 September 2026

A reliability release. No changes to gene sets, analyses or the interface — the application code is identical to 1.6.0.

Changed

  • The site now serves several requests at once per server process. It had been serving exactly one, which meant a single slow page could make the whole site unresponsive to everyone else. Capacity at the minimum deployment size went from two concurrent requests to eight.
  • A request that stalls is now abandoned after 60 seconds instead of 300, so a stuck page releases its worker in a minute rather than five.
  • Health checks were added. A server that is saturated or wedged is taken out of rotation automatically, and one that does not recover is restarted. Readiness now also depends on the search service being ready, so a server cannot accept searches before its index is available.
  • Every outbound call now has a time limit. Searches wait at most two seconds for the search service, and PubMed and SRA lookups are bounded. An SRA lookup that times out or is rate-limited now leaves the publication metadata blank instead of failing the gene set page.

1.6.0 — 15 September 2026

The first release built from the consolidated GeneWeaver repository, replacing 1.5.27. It gathers roughly a year of fixes, including work that had been merged but never reached geneweaver.org, so it is much larger than a normal release. The pre-releases 1.6.0a, 1.6.0b and 1.6.0c are included here.

Gene set membership and gene counts changed

Some gene sets are a different size after this release

Three long-standing defects meant stored membership did not match the threshold rules the interface described. Correcting them changed which genes are in threshold on existing, published gene sets — so analysis results on those sets change too. This is the correction landing, not new damage. If a set of yours looks different, the explanations below are the likely reason.

  • Binary gene sets were unusable and now work. A binary (membership) set records presence with a value of 1, but the database was testing value > 1, so every gene in such a set was marked out of threshold. The sets existed, looked populated on their own page, and returned nothing in every analysis tool. Whole sets went from unusable to usable.
  • Correlation and Effect thresholds now use the signed value, not its magnitude. A range such as 6.0 < Effect < 22.5 was also admitting values between −22.5 and −6.0, because the comparison discarded the sign. Sets with an asymmetric range get smaller where negative values were wrongly included — and, where the range is itself negative, larger, because a magnitude can never fall inside a negative range. Symmetric ranges are unaffected.
  • Gene counts in search results and My GeneSets are now derived from the genes actually stored. Three separate upload paths recorded a count taken from something other than the saved genes — submitted lines including a trailing blank one, staged rows before duplicate identifiers were merged, or one row per identifier rather than per gene. Search and the gene set page therefore disagreed, always with search over-counting. The count is now computed after the genes are saved, which makes it honest rather than larger: where identifiers failed to resolve, the number shown goes down.

Fixed — curation and upload

  • Ontology annotations could not be generated at all, broken since around June 2025. Annotation also now offers only the ontologies that are actually supported; the non-functional Monarch and "Both" options are hidden.
  • Genes were silently dropped on upload. An identifier that was an alias or synonym rather than a gene's official symbol was discarded with no warning, so an uploaded set was quietly smaller than the list submitted.
  • Score values are now validated on upload, on both single and batch upload, instead of out-of-range or invalid values being accepted silently. A gene set's score type can also be changed after upload.
  • A batch P-Value or Q-Value threshold could never take effect. A header such as P-Value < 0.05 was parsed as a whole line where a bare number was expected, so every thresholded line fell through and the cutoff was silently replaced with the default 0.05. Malformed variants (P-Value > 0.01, P-Value 0.01) were treated as a deliberate default and also replaced the cutoff with no warning; they are now reported.
  • Changing a score type from Correlation or Effect to P-Value or Q-Value lost the change. The two-sided types hold a low,high pair and the one-sided types hold a single number, and saving a pair under a one-sided type failed with only "An unknown error ocurred." The threshold field now reshapes when the score type changes, the previous value is remembered per score type so switching back restores it, and where a default has to be substituted you are told.
  • Changing a score type now checks the values against it, warning — not blocking — when existing values make no sense under the new type, matching what upload already did.
  • Creating a gene set from a tool result silently dropped genes. The gene list was resolved through the homology table, so any gene absent from Homologene could not survive — even when transposing within a single species, where nothing needs transposing. Measured on one real set: 6 of 115 genes lost, with no warning anywhere. Unchanged since April 2018.
  • Gene sets created by analysis tools skipped threshold processing entirely, carrying membership from a hardcoded rule that matched no score type and was applied before the score type was known. Both creation routes now use the same threshold logic as every other path.
  • Changing a gene set's tier from the admin page had no effect, and some gene sets returned an error when opening the edit view.
  • Search results were only as fresh as the last server restart. The search index was built once when a server started and never again, so a new or edited gene set could be missing from search for weeks. The index is now refreshed every 15 minutes with a full rebuild nightly. Saving an edit also updates the gene set's modification time, which it previously did only when the edit page was opened — so an edit saved from a long-open page could never be picked up.
  • Zero-result and unsorted searches returned an error instead of an empty result page.
  • Clearing every option in a filter returned nothing at all, rather than meaning "no restriction".
  • Find Similar GeneSets compared a filtered set against unfiltered candidates, inflating similarity scores. On a sample of 300 candidate sets, 67 were affected.
  • Find Similar GeneSets and the Jaccard Similarity tool now agree. Genes with no homology entry were dropped by one and kept by the other.

Fixed — analysis tools

  • Boolean Algebra's Symmetric Difference returned an error. A diagram failure also no longer blanks the whole visualisation when comparing three or more sets.
  • MSET reported "list not a subset of its background" for valid input, after a gene data reload left the backgrounds stale.
  • MSET gave a cryptic server error for Tier IV gene sets. Such a set can contain real genes that fall outside the curated background, which MSET correctly cannot use. You now get a readable message naming how many genes are outside the background and which they are, instead of a generic error. The underlying limitation is unchanged and deliberate.
  • MSET internal errors from an incomplete Python 3 migration, and a worker crash that used the wrong background for the second list.
  • DBSCAN crashed when a run produced no clusters. A run with no clusters is now a result, not an error.
  • Diagram-producing tools failed when the graph renderer was not at an expected path.
  • Jaccard Similarity reported a p-value of 0 for any pair of gene set sizes not already in its sampling cache, and the caches were missing in some environments.

Security

Both of the following affect earlier versions. If you run your own GeneWeaver instance, they are the reason to upgrade.

  • The admin data-table endpoints accepted SQL through request parameters and ran it unbound, which allowed statements beyond reading — and two of the four endpoints turned out not to require an administrator, or any login, at all. The same two also took the user whose gene sets and tool results to list from the request, so anyone could list another user's by guessing an identifier. All parameters are now bound, table and column names are checked against a fixed list, and the user is taken from the session.
  • An authentication client secret was written to the server log on every start-up. It no longer is. Anyone self-hosting a version before 1.6.0 should treat that secret as exposed and rotate it.

Added

  • Batch upload accepts pasted text, not only a file.

Changed — where a value exactly on its threshold belongs

For P-Value and Q-Value gene sets, a value exactly equal to the cutoff is outside the threshold. Two parts of the application had disagreed about this for years, so a value on its cutoff could be a member or not depending on which one last saved the set.

The behaviour every environment has actually stored was the exclusive rule, and it was kept: no gene set's membership was changed by this decision. Choosing the other reading would have retroactively added genes to roughly 600 published gene sets per environment, some created as far back as 2007, which is a change to the scientific record rather than a fix. The Set Threshold preview count, which had used the inclusive rule and could therefore promise more genes than the set would end up holding, now matches. Correlation and Effect ranges remain inclusive at both ends, which was never in question.

Known issues

  • Some gene sets claim genes while holding none. A number of older gene sets record a gene count but contain no genes. The count was deliberately left alone rather than reset to zero: where genes may have been lost, a stale number is a better record than an authoritative zero. Being investigated separately.
  • MSET cannot run against gene sets whose genes fall outside the curated background. The error message is now clear about why; the limitation is unchanged, and is expected to be resolved by the next generation of the analysis tools.
  • A threshold of NaN puts every gene in threshold. The database accepts the spelling and treats it as greater than any number. Correcting it would move published membership, so it is being handled as a curation decision rather than a defect fix. No live gene set currently has one.
  • "Find other GeneSets from this publication" fails on very large publications. For a publication with thousands of associated gene sets the page does not finish loading, and for a gene set with no publication it returns an error. Long-standing — present in every earlier version. Fixed in 1.6.2.