Several follow-ups block on a multi-sample entry point that monitor (single patient) and estimate-bias (single pair) do not provide. Build the shared entry point once; the payloads below consume it.
Shared piece: allomix cohort entry point. Manifest-driven. Each row: role (host/donor/admix), the patient/pair it belongs to, the VCF path, the ##allomixRunUnit. Reads the same per-sample informative markers + admix allele counts as monitor, plus the per-sample contamination estimate and run-unit metadata.
Payload 1 (highest value, [data]-backed): cohort-recurrence bad-site blacklist.
A panel-level blacklist for loci systematically inconsistent across the repeated cohort. Evidence from SRP434573 (same 7 people across ~64 runs, so a bad locus is bad in every sample): ~44 site-instances across the 10 two-person mixtures sat at ~42% minor allele against a ~0.004% error floor and ~0.2% contamination, carrying ~92% of the pooled minor reads. Emit a blacklist monitor can consume panel-wide. Check against the clean-control work so it does not blacklist good loci.
Follow-on payloads (smaller, also need the entry point):
- Batch / run-level contamination QC
[likely]: group admix samples by ##allomixRunUnit and flag a whole flowcell lane when its samples share an elevated floor. Contamination differed several-fold between the M3-F3 (v2) and F2-M1 (v1) libraries; same-run + cohort-wide elevated floor is stronger index-hopping evidence than the run-unit provenance flag or the in-data estimate alone.
estimate-bias --both-het cohort entry point [likely]: a pair's both-het markers only help other pairs, so the pooled table (the only route to the homozygous-in-both fully-informative markers) needs the multi-sample entry point. A pair's table covered 0 of its own 576 informative markers but 66% of a held-out pair's informative set when pooled across 11 mixtures. Refinement (small interior gain, neutral at low fractions), not a sensitivity lever.
Evidence: Payload 1 [data]/[likely]. Was Step 33 in the retired overall plan.
Several follow-ups block on a multi-sample entry point that
monitor(single patient) andestimate-bias(single pair) do not provide. Build the shared entry point once; the payloads below consume it.Shared piece:
allomix cohortentry point. Manifest-driven. Each row: role (host/donor/admix), the patient/pair it belongs to, the VCF path, the##allomixRunUnit. Reads the same per-sample informative markers + admix allele counts asmonitor, plus the per-sample contamination estimate and run-unit metadata.Payload 1 (highest value,
[data]-backed): cohort-recurrence bad-site blacklist.A panel-level blacklist for loci systematically inconsistent across the repeated cohort. Evidence from SRP434573 (same 7 people across ~64 runs, so a bad locus is bad in every sample): ~44 site-instances across the 10 two-person mixtures sat at ~42% minor allele against a ~0.004% error floor and ~0.2% contamination, carrying ~92% of the pooled minor reads. Emit a blacklist
monitorcan consume panel-wide. Check against the clean-control work so it does not blacklist good loci.Follow-on payloads (smaller, also need the entry point):
[likely]: group admix samples by##allomixRunUnitand flag a whole flowcell lane when its samples share an elevated floor. Contamination differed several-fold between the M3-F3 (v2) and F2-M1 (v1) libraries; same-run + cohort-wide elevated floor is stronger index-hopping evidence than the run-unit provenance flag or the in-data estimate alone.estimate-bias --both-hetcohort entry point[likely]: a pair's both-het markers only help other pairs, so the pooled table (the only route to the homozygous-in-both fully-informative markers) needs the multi-sample entry point. A pair's table covered 0 of its own 576 informative markers but 66% of a held-out pair's informative set when pooled across 11 mixtures. Refinement (small interior gain, neutral at low fractions), not a sensitivity lever.Evidence: Payload 1
[data]/[likely]. Was Step 33 in the retired overall plan.