Skip to content

Somatic pipelines

Somatic small-variant pipelines validate against the SEQC2 HCC1395 / HCC1395BL pair — a breast-carcinoma cell line and its matched B-lymphoblastoid line, both public ATCC reference materials. Aver compares your pipeline's somatic VCF against the SEQC2 high-confidence sSNV and sINDEL call sets (release v1.2.1, GRCh38) inside the published high-confidence regions, using som.py from the same digest-pinned comparator image as every other comparison.

Everything downstream is the same product: metrics per run, drift when the pipeline's configuration changes, baselines, and signed evidence packs.

The flow

# Run your somatic pipeline on the HCC1395/HCC1395BL pair yourself, then:
aver compare --query hcc1395.somatic.vcf.gz --sample HCC1395 \
  --region chr20 --snapshot snapshot.json \
  --output aver-out/somatic-hcc1395

aver push --results aver-out
  • Truth files are fetched from NCBI and verified against recorded SHA-256 checksums before use; the first run downloads them once.
  • som.py runs once per class (sSNV, then sINDEL); each class's row comes from its own truth set, and both command lines are recorded on the run.
  • Counts follow som.py's defaults: PASS query records, with calls outside the high-confidence regions counted as unknown rather than false positives.

Sample naming (strict, by design)

Aver only processes VCFs it can prove belong to a public reference material. For somatic comparisons, every sample in the VCF (one or two) must be named with the registered names of the pair — HCC1395 and HCC1395BL. Caller defaults like TUMOR/NORMAL are rejected, because a generic name proves nothing. Rename once at export time:

bcftools reheader -s <(printf 'TUMOR HCC1395\nNORMAL HCC1395BL\n') \
  -o renamed.vcf.gz your.somatic.vcf.gz

The error message never echoes an unrecognised sample name.

What the limits of this are

  • One contig per comparison for now (--region chr20), matching the germline flow.
  • Stratifications and VAF-binned reporting are not yet offered for somatic runs; nothing is estimated outside the comparison tool.
  • Draft truth sets (such as GIAB's somatic HG008) are not used; only the published SEQC2 release is catalogued, and the catalogued version with its checksums is recorded on every run.