HLA typing

Across about 3.6 megabases lie more than 200 genes, many with immune function. The classical HLA genes are extremely polymorphic: for HLA-B several thousand alleles are known. This diversity is an evolutionary advantage against pathogens, but it makes typing technically demanding. Tightly neighbouring loci, high sequence similarity between alleles and gene conversion further complicate unambiguous assignment.

Genome uses HLA*LA (Dilthey et al., 2019). Instead of a linear reference it relies on a population reference graph (PRG) of the MHC that directly encodes known IMGT/HLA alleles of all classical loci as alternative paths. Reads from the MHC portion of the alignment are projected onto this graph, so even strongly divergent alleles can be assigned that would map poorly against a single linear reference. From the projected read paths HLA*LA infers the most likely allele combination per locus. The chr1 and chr2 columns are HLA*LA output columns, not a phasing claim, because the calls are not phased across different loci.

Genome interprets the main R1_bestguess_G.txt output as G-group output: class I calls refer to exons 2 and 3 and class II calls to exon 2. A G group collapses alleles that are identical across the antigen-relevant exons, so several compatible IMGT/HLA alleles can apply to the same G group; Genome then shows the first entry as the top candidate and lists further compatible candidates in the appendix. The resolution usually reported reaches the two-field level (allele family and protein), whereas deeper fields for synonymous or non-coding differences often cannot be resolved from short-read WGS. As a quality value, Q1 is the central HLA*LA score and is typically close to 1; Q2 is not treated as a report criterion according to the HLA*LA documentation and is carried only as a raw metric. A perfectG=1 flag indicates a perfect translation of the internal call into G-group resolution; if perfectG is not 1, the untranslated R1_bestguess.txt output should also be reviewed.

Short-read WGS yields short fragments that often cannot unambiguously separate individual alleles, because the distinguishing positions lie farther apart than the read length. Within a G group the antigen-relevant exons are identical, so several alleles explain the same reads and can only be reported as a compatible set. In addition, two haplotypes are present at once and the graph must keep both paths apart, which can fail for similar alleles with uneven coverage. Genome therefore reports coverage and plausibility metrics alongside the top candidate and flags calls as unique, ambiguous or not typed. For transplantation-relevant decisions, confirmatory laboratory typing remains the standard.

Individual HLA alleles are tightly linked to disease, for example HLA-DRB1*15:01 with multiple sclerosis, HLA-B*27 with ankylosing spondylitis, or HLA-B*57:01 with abacavir hypersensitivity. Genome presents the typing as technical evidence and separates that evidence from interpretation. It does not replace qualified medical assessment or accredited laboratory typing.