Repository navigation
Ogataea routing: a lineage-routed MATog family with all hits withheld leaves no fallback to MATsc/MTL #228
Description
Finding
In the v0.8.0 run, three Ascomycota loci that v0.7.0 called are now uncalled in genomes routed by lineage to MATog (Ogataea).
| Genome | v0.7.0 call | v0.8.0 |
|---|---|---|
GCA_019232045.1 (O. degrootiae) JAHDIT010000187.1:14392-15783 |
MTL alpha, high (MTLalpha1, MTLalpha2) |
MATog alpha withheld (modelled_gene_bar); no call |
GCA_030557025.1 JAKTVC010000474.1:52-1443 |
MTL alpha, high (MTLalpha1, MTLalpha2) |
MATog alpha withheld (modelled_gene_bar); no call |
GCA_003707685.2 PPKM02000116.1:542-706 |
MATsc undetermined, medium (MATA2, MATALPHA2) |
MATog alpha withheld (modelled_gene_bar); no call |
Cause
v0.7.0 searched these genomes by phylum fallback with MATsc and MTL. In v0.8.0 the Ogataea genus (461281) is in the MATog scope, so routing is by lineage and only MATog is attempted (checked on GCA_019232045.1: families_attempted changes from MATsc, MTL to MATog). When the MATog hits are withheld by the modelled-gene bar there is no fallback. A rerun of the frozen commit with --preset v0.7 (cross-family policy off) gives the same result, so the cross-family policy is not the cause.
Proposed fix (v0.8.1)
When a lineage-routed family has all its hits withheld, also search the phylum-fallback families (MATsc, MTL) and report their calls (flagged out-of-scope where the family rules require it), or admit the MATog fragments at low confidence.
Status
Accepted as a known issue for v0.8.0 (curator, 2026-11-02); the definitive run is not re-run for it. Listed under Known issues in the v0.8.0 changelog. Data: results/2026-10-31_v080_submit/diff_vs_v070/ascomycota/regression_diff.tsv (call_lost rows with no overlapping call from another family).
🤖 Generated with Claude Code
Activity
Changelog: link the Ogataea routing known issue to #228
Release prep v0.8.0 (draft; do not merge until the release is cut) (#207
Changelog: fallback after withheld (#228)
Detection: search the phylum-fallback families when a routed family's…
hyphaltip commentedon Nov 5, 2026
Moved to v0.9.0 (curator, 2026-11-05). The fix of #253 (
--fallback-after-withheld-policy on, presetv0.8.1) is in v0.8.1 and recovers one of the three Ogataea genomes it was tested on. Remaining: genomes where one cassette is called and another locus is withheld by the modelled-gene bar (the 56 one-cassette Ogataea genomes); widening the fallback does not reach them because the right family already called, so the open question is the withheld loci themselves (hide-one-cassette test on the 71 two-cassette genomes, see #256).🤖 Generated with Claude Code
hyphaltip commentedon Nov 6, 2026
Closing as implemented (curator's decision, 2026-11-06).
The behaviour is
--fallback-after-withheld-policy on(off by default in the code; on in thev0.8.1preset used by the definitive run). In the v0.8.1 run 17 genomes are called that were not before: 7 Ascomycota (Naumovozyma, Ogataea, Babjeviella, and one each of Terfezia, Monosporascus, Biscogniauxia, Ascodesmis) and 10 Basidiomycota (15 loci).--preset v0.8reproduces the v0.8.0 reports for the 7 Ascomycota genomes, so the effect is exactly the option, and no call was lost elsewhere (PR #277, release v0.8.1).Whether the option should become the code default is a separate decision for v0.9.0; open a new issue if it is wanted.
🤖 Generated with Claude Code