HaplotypeCaller

HaplotypeCaller is a variant-calling tool in the Genome Analysis Toolkit (GATK) developed by the Broad Institute. It detects single nucleotide variants (SNVs) and small insertions and deletions (indels) from mapped sequencing reads by reconstructing possible haplotypes in local genomic regions and evaluating their likelihood given the observed read data.

The algorithm identifies "active regions" with evidence of variation, performs local de novo assembly (using de Bruijn-like graphs) to generate candidate haplotypes, and computes haplotype likelihoods with a pair hidden Markov model (pairHMM). Candidate haplotypes are aligned against the reference to call variants and estimate genotype likelihoods. HaplotypeCaller can operate in per-sample mode and in GVCF mode, which emits reference confidence records to support scalable joint genotyping across cohorts; joint genotyping is typically done with the GenotypeGVCFs tool.

HaplotypeCaller is widely used in human and non-human genomics due to its improved handling of indels and complex regions compared with simple pileup-based callers. It is implemented in Java and distributed as part of GATK, with major updates in GATK4 emphasizing performance and runtime improvements.

Limitations include increased computational and memory demands relative to lightweight callers, potential sensitivity to read mapping quality and sequencing artifacts, and the need for downstream filtering to reduce false positives. Users commonly follow best-practice workflows for data preprocessing, variant quality recalibration or hard filtering, and joint genotyping for cohort studies.