Nasir Nesirli

Clinical genomics pipelines, from raw reads to benchmarked variants and finished reports. Ten years in clinical labs before that.

Repository: nesirli/germline-wgs-giab · Status: in progress. Results will be added here when the benchmark is complete.

Why it matters

Germline variant calling is the core of every genetics diagnostic lab. Validating a pipeline the way ISO 15189 expects means measuring accuracy against reference materials, broken down by genomic context, not just quoting one overall F1.

Benchmark design

  • Truth: GIAB v4.2.1 small-variant benchmarks for HG001–HG007 (GRCh38), with GIAB genome stratifications (low mappability, segmental duplications, homopolymers, tandem repeats, MHC, GC extremes) and the HG002 T2T-Q100 benchmark for harder regions.
  • Reads: GIAB Illumina 2×150 PCR-free WGS. HG002 as a full genome (~30×), the other six on chr20.
  • Metrics: SNV and indel precision, recall and F1 per sample and per caller, stratified by region type, plus runtime and cost per sample.

Pipeline

fastp → BWA-MEM2 → MarkDuplicates and BQSR → QC (mosdepth, VerifyBamID2, somalier sex check, Ti/Tv) → GATK HaplotypeCaller and DeepVariant → hap.py against GIAB truth → VEP with gnomAD, ClinVar, REVEL and SpliceAI → Exomiser with HPO terms → semi-automatic ACMG/AMP evidence → clinical-style PDF report.

Demo case

HG002 with a heterozygous LDLR variant spiked in silico (fictional familial hypercholesterolaemia case). The report states that the data are public and the patient details are fictional.

Updated: