STRC Research

pr84-body

3 min read

Summary

  • Classify STRC-family haplotypes from both configured identity signals: the pivot SNV and the 314 bp STRCP1 deletion.
  • Label only agreeing marker pairs as STRC or STRCP1; label discordant, missing, truncated, or internally inconsistent evidence as unknownhap.
  • Report gene_cn: null when any haplotype identity is unknown while retaining independently supported total_cn.
  • Omit unknown haplotypes from the default locus-specific VCF instead of projecting them onto STRC.
  • Add a public HG005 truth fixture and integration test for the independently established 3 STRC + 1 STRCP1 copy structure.

This PR is based directly on main. The complementary regression-only PR #83 covers Python-to-Rust parity on HG00733 but is not a dependency.

Problem

The STRC caller previously classified any haplotype containing deletion category 3 as STRCP1 and every other haplotype as STRC. That forced missing deletion evidence into STRC and forced discordant pivot/deletion pairs into one of the two loci. The VCF split had the same fallback: every name not containing strcp1 went to the STRC reference.

This change makes uncertainty explicit rather than turning incomplete evidence into a locus assignment.

Behavior

Pivot314 bp deletionOutput
reference (1)absentSTRC
alternate (2)present (3)STRCP1
reference (1)present (3)unknownhap
alternate (2)absentunknownhap
missing, truncated, or inconsistentanyunknownhap

The deletion evidence reader supports both representations already produced by the phaser: a synthetic deletion site or category 3 spread across heterozygous sites inside the deletion interval. Mixed deletion/non-deletion evidence is unresolved.

An unknownhap is an abstention, not a gene-conversion or hybrid call. The default two-reference VCF omits it. Explicit --gene1only retains its existing behavior of projecting all named haplotypes onto the main-gene reference.

Independent HG005 gate

The new 524-read locus fixture comes from the public NIST GIAB HG005 Sequel II CCS alignment. The expected locus structure is supported by MLPA and divergent-marker refinement in the Tsai et al. preprint; the source is not peer reviewed, so this is a locus/CN benchmark rather than clinical validation.

The integration test reproduces:

  • total CN 4
  • STRC CN 3
  • STRCP1 CN 1
  • region depth 81/85 and intergenic depth 34
  • NM_153700.2:c.5125A>G (chr15:43600074 T>C) on exactly one STRC copy
  • agreeing (1,1) STRC and (2,3) STRCP1 identity-marker pairs

Validation

HG38=<GRCh38-no-alt.fasta> cargo test --workspace --all-targets
cargo test: 145 passed (6 suites, 40.74s)
 
cargo fmt --all -- --check
git diff --check
samtools quickcheck -v tests/data/bams/HG005.strc.bam

Focused unit coverage includes agreeing, discordant, missing, truncated, synthetic-deletion, internal-deletion, mixed-evidence, copy-counting, and VCF-routing cases.

I also cloned the published head commit 0063b992032e5bb0f297a492d43683b5005309e7 into a detached clean worktree, built it in release mode, and ran the full HG005 CLI gate twice. Both calls were byte-identical (JSON SHA-256 1b6fc27923f611f74dc0f02c28f67ebf1054c9971be611d4759adb003e1d073b) and passed every locus/CN check. The fixture interval hash also matches the depth probes embedded at this exact commit.

cargo clippy --workspace --all-targets completes with no errors and the repository’s existing 69 warnings; none point to the added STRC classifier, VCF routing helper, or HG005 test.

Provenance

  • Source BAM: https://ftp.ncbi.nlm.nih.gov/ReferenceSamples/giab/data/ChineseTrio/HG005_NA24631_son/PacBio_SequelII_CCS_11kb/HG005_GRCh38/HG005_GRCh38.haplotag.10x.bam
  • Official source BAM MD5: ec0bbf687f0fd5d0053cbeb761111deb
  • Official BAI MD5, locally verified: b1c28c3258e4f0f097ff6a4c6fa3ed43
  • Truth source: https://doi.org/10.1101/2024.11.05.24316795
  • Reference: GCA_000001405.15 GRCh38 no-alt analysis set
  • Reference FASTA MD5: 30f1a99359a5993edbf8c8bc885825a7
  • Fixture reads: 524
  • Fixture BAM SHA-256: 030e1e35b2c6d8c11b07c0580bb1aa929d199c0026f83b1e67a7bca9860efc6e
  • Fixture BAI SHA-256: 8afb34b8910163db60367f815ac6ddffb0552d39234e344ac402cfe93350c446

Scope

This establishes conservative locus identity and copy-number behavior on public HiFi data. It does not establish pathogenicity, clinically validate Paraphase, or claim that every unknownhap is a conversion event.

Graph View