Repository navigation
Conversation
…ectory Document the 0.9 heterozygous FORMAT/AF threshold and the DeepVariant fallback. Classify from the Nextflow task directory and test threshold boundaries offline. Bead: lin-37 Bead: lin-40 Bead: D10
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 33 minutes. View limit details
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. |
|
Local regression evidence for e7fdd67. The pinned image The classifier launcher was wrapped with a guard that required its working directory to stay under The task scripts came from the The ordinary smoke script also passed in the same read-only image. Its new synthetic checks run in the existing HAPLOGREP3_IMAGE smoke row in Container Test. A second offline/read-only run used the same five synthetic GT:AF calls for both tasks. The legacy task ran from the binary directory with the default threshold. The fixed task ran from its task directory with explicit 0.9. The local personal-data guard also passed: |
…rding # Conflicts: # CHANGELOG.md
Step 12 said that haplogrep3 uses heteroplasmies, but its VCF profile excludes heterozygous FORMAT/AF calls below 0.9. The Nextflow task also changed into the binary directory, based on an old reason about downloading config.
Both callers now pass
--hetLevel 0.9explicitly. The docs explain the AF threshold, the FASTA's separate 0.5 rule and DeepVariant's unread FORMAT/VAF. The Nextflow task stays in its work directory and uses the pinned image's bundled config and trees offline. No output name, format or classification threshold changes.Decision: keep the approved 0.9 threshold. Changing it to 0.5 could change haplogroup assignments and is outside this fix.
How it was checked
--network none --read-only, an empty writable task directory and a read-only binary directory. Synthetic markers at AF 0.90 and 0.95 were retained. AF 0.05 and 0.60 were excluded. A homozygous ALT marker stayed despite low AF.--hetLevel 0.5failed the profile assertion with exit 1. These used synthetic input only. The whole harness ran again on head 4993e74 (only a main merge into CHANGELOG.md since e309d01): main's task still fails the directory guard (FAIL: classification left the task directory, exit 97), the fixed task passes, and the--hetLevel 0.5copy still fails (FAIL: haplogrep3 profile does not obey hetLevel 0.9, exit 1). The guard wrapper sits in a tmpfs, so that mount needsexec.cmp) on the synthetic threshold fixture under the same offline/read-only conditions.scanned 5 text files.shellcheck -S warningand shell syntax checks passed for both shell files.git diff --checkpassed.ORACLE_RUNNER=local scripts/ci/oracle-diff.sh --self-test: all 15 checks passed, including 44 comparator cases. No Rust tool changes, so no tool-specific oracle cases apply.Known gap: no CI check reads the
MITO_HAPLOGROUPscript block itself. The smoke row runs its own copy of the classify command. A static guard (fail if the script block changes directory) needs a file outside this change, so it is left as a follow-up.Part of #411. Settles lin-37, lin-40 and owner decision D10.