Propagation by Homology

Bottom line: many GO annotations reach a gene by transfer from another gene
product: an ortholog (ISO), a sequence a curator judged similar (ISS/ISA), a
PANTHER family node placed by a PAINT curator (IBA), or an electronic pipeline
that does the same automatically (Ensembl Compara, TreeGrafter, InterPro2GO).
A transfer can go wrong in three places: the donor annotation can be weak or
stale, the orthology or tree relation can point to the wrong gene, or the term
can describe a context that does not carry over to the target. This page indexes
the projects that review these transfers method by method, and a propagation
browser that shows each transferred annotation as donor, intermediate and target
beside the gene review's verdict. We built it so that reviewers record where each
defect sits, and so that a fix can go to the mapping, node or pipeline that
produced it, where one change corrects every affected gene. In the statistics
generated on 2026-09-26, reviewers had judged 34,253 of 34,265 propagated
annotations and flagged 13% as REMOVE, MARK_AS_OVER_ANNOTATED or MODIFY: 9% of
PAINT IBA, 20% of Ensembl Compara and 28% of TreeGrafter annotations. Of the
4,232 ISO annotations, 383 rest on a donor that no longer carries the term, and
3,000 (71%) add an assertion with no related IBA on the target.

Browse all propagated annotations — one row per
transferred GOA annotation, read as donor(s) → intermediate → target, with what
the donor carries today, what else the target already carries, and the gene
review's verdict. Filter by method, donor or target species, donor support,
review action, or failure mode; the URL records the selection.
Browser guide ·
Current statistics

The three parts of every transfer

Every propagated annotation has the same shape, whatever the method:

A review therefore asks three separate questions, and a failure can sit in any one
of them:

1. Is the donor annotation sound? Does the donor still carry the term, with
   experimental support, from a paper about that gene?
2. Is the relation the right one? One-to-one ortholog, a paralog, one of an
   expanded family, or a node that the target should not descend from.
3. Is this term safe to move? Conserved activity versus tissue, lineage,
   compartment, or regulatory context that does not transfer.

The shared vocabulary for recording the answer is the
ISO/IBA failure taxonomy (review.propagation_review).

Projects

Status and follow-ups

The propagation browser and current statistics are built and committed from a
2026-09-26 donor cache. Component projects continue to move independently: PAINT
no-IBA reviews, the IBA corpus re-review, InterPro2GO and NCBIFAM/CDD mapping
proposals, TreeGrafter recomputation, and the ortholog-conjecture pilot each
need their own refresh or curator-facing handoff before the next umbrella
snapshot. The cross-project tracker is
#4095.

Methods covered by the browser

Method definitions follow the GO_REF descriptions in
go-site gorefs.yaml.

The browser also includes ISO/ISS/ISA rows that cite a paper instead of a GO_REF.
ISM (sequence-model predictors such as AtSubP) is not homology transfer and is not
included.

Rebuilding

just refresh-propagation-sources   # network: donor identities (UniProt), donor GO (QuickGO), GO closure
just deploy-propagation-browser    # offline: app/propagation/
just propagation-stats             # offline: HOMOLOGY_PROPAGATION/propagation-stats.md

The donor cache lives in HOMOLOGY_PROPAGATION/data/ and is committed, so the
browser and statistics rebuild offline and reproducibly; refresh it when GOA
files are re-fetched.