The problem. You can run GATK and get a VCF, but how do you know it’s correct? Variant calling has no ground truth in a typical sample — you can’t independently confirm millions of calls. Without a trusted reference set of high-confidence variants and the regions where you can trust them, benchmarking a caller is guesswork.

The idea. Genome in a Bottle sequenced reference samples (notably HG002 and family) with many technologies and integrated them into a high-confidence truth set: a curated list of variants plus the genomic regions where calls can be trusted. Compare your caller’s output to this inside those regions and you get real precision/recall numbers.

Why it matters. This is how I’d validate variant_calling_nf honestly. GATK produces calls; GIAB tells me how many are right. Reading it reframes variant calling from “run the tool” to “measure the tool against a standard,” which is the difference between using a pipeline and trusting it. It pairs directly with the Platinum Genomes benchmark alongside it today.

Verdict. Foundational infrastructure for the whole variant-calling field — GIAB truth sets underpin nearly every caller benchmark. Read it for the confident-regions idea: knowing where you can trust a call matters as much as the calls themselves.