Phasing
The tool LightSpeed Long Reads to Germline Variants (beta) adds phasing information to variants, provided read evidence supports it.
Phasing is performed in an explicit step after variant calling. The phasing step uses the called variants together with the supporting read alignments to determine which alleles at nearby heterozygous sites are present on the same chromosome copy, and groups such variants into phase sets.
The strategy used for phasing is derived from the WhatsHap phasing algorithm [Patterson et al., 2015]. For each chromosome, heterozygous variants are used to build a fragment matrix describing the alleles observed in reads spanning multiple variant sites. Candidate phase blocks are then identified from connected groups of variants linked by read evidence. For each such block, phasing is determined by solving a Minimum Error Correction (MEC) problem, minimizing disagreement between observed read alleles and inferred haplotypes.
When phasing, all SNVs are considered. Indels are considered if they have QUAL at least 99. Indels with lower QUAL are only considered if the the germline variant caller's internal phasing considerations suggest that they can be phased.
To limit the computational complexity, informative reads are selected before phasing, while preserving reads that connect otherwise separate blocks. Homozygous alternate variants can subsequently be attached to overlapping phase sets when supported by the allele counts.
The phasing information is available in the genotype track, see Genotype track.