Skip to content

Structural variants

Pipelines that call structural variants (deletions, insertions and other events ≥50bp) validate against the GIAB HG002 SV Tier 1 v0.6 truth set, compared with truvari exactly as the truth set's own README recommends (--passonly, the Tier 1 regions BED, -r 2000), from a container pinned by digest.

GRCh37 only, by upstream design

The v0.6 Tier 1 release is published on GRCh37; GIAB states its GRCh38 SV callsets are under development. Aver therefore requires --build GRCh37 explicitly and refuses anything else with an explanation, rather than lifting coordinates itself.

Usage

aver compare --query hg002.sv.vcf.gz --sample HG002 --sv --build GRCh37 \
  --snapshot snapshot.json --output aver-out/sv-hg002

aver push --results aver-out
  • Truth files (VCF, index and Tier 1 BED) are fetched from NCBI and verified against recorded SHA-256 checksums before use.
  • The comparison runs genome-wide within the Tier 1 regions; a plain-text VCF query is bgzipped and indexed for you in the pinned bcftools container.
  • Results carry one SV metric row — recall, precision, F1 and the TP/FN/FP counts straight from truvari's summary.json, with every other field it reported preserved on the run.
  • For callsets that are not sequence-resolved, --sv-ignore-sequence matches on size and type only (truvari --pctseq 0, the README's own suggestion).

The usual rules hold: the query VCF's sample must be HG002 (or a registered alias), and the truth set's version and checksums are recorded on every run.