HLA*LA from WGS: how closely do calls match laboratory typing?
A direct comparison with targeted HLA sequencing found about 94 percent allele agreement. Results vary by HLA locus, and the figure applies to a defined benchmark.
HLA*LA can infer HLA types from aligned whole-genome sequencing data. How often do those calls match targeted HLA laboratory typing? A multi-platform comparison published as a 2026 preprint provides a concrete figure: about 94 percent in the direct assay comparison. That result applies to a specific study design; it is not a blanket guarantee for every sample.
The direct comparison: 93.7 to 94.5 percent
Cremin et al. compared HLA*LA calls from WGS with GenDX NGSengine. The authors describe GenDX as a clinically validated sequencing assay: it amplifies the HLA region by PCR and then sequences it on Illumina. The direct laboratory comparison covered 24 Coriell cell lines. The full study included 29 specimens, but the GenDX reference type was not available for all 29.
The comparison covered eight shared loci: HLA-A, -B, -C, -DPA1, -DPB1, -DRB1, -DQB1 and -DQA1. HLA*LA was run on WGS data from Illumina, MGI, and Oxford Nanopore R9 and R10. Most datasets had roughly 30-fold coverage. The study counted correctly matched alleles per gene, comparing calls without regard to allele order at two-field resolution.
| Direct comparison with GenDX | Result |
|---|---|
| Eight shared HLA loci, mean for each WGS platform | 93.7–94.5% |
| HLA-A, -B and -C | above 98% each |
| HLA-DQA1 and -DQB1 | about 83–85% on average |
The roughly 94 percent figure is an allele-level match across the tested loci and platforms. It hides differences between genes: the class I loci scored higher, while DQA1 and DQB1 were more difficult to type. The preprint abstract summarizes results at about 95 percent across 13 HLA genes. For the specific question of agreement with a laboratory assay, however, the narrower comparison of 24 cell lines and eight shared loci is the relevant one.
What the other percentages measure
The study also examined blood, buccal swabs and saliva. Reported concordance was 98.5–99.2 percent for blood, 93.1–93.8 percent for buccal swabs and 81.5–85.4 percent for saliva. These are not agreement rates against an independent laboratory reference: orthogonal reference types were unavailable for those samples. Instead, the authors compared HLA calls from the same individuals across sample sources and sequencing platforms.
The coverage analysis is different as well. At 15-fold coverage, 13 of 20 sample-platform combinations matched their corresponding 30-fold calls exactly; all were above 90 percent concordance. The reference was each sample’s own HLA*LA call at 30-fold coverage, not GenDX. These numbers show internal consistency at lower coverage, not 15× agreement with a laboratory test.
How other studies fit
A peer-reviewed 2024 study compared HLA calls for 36 samples previously typed by targeted NGS. HLALA was selected for HLA-B, -C, -DPB1 and -DQB1; a combined tool strategy performed better for HLA-A and -DRB1. This supports useful but locus-dependent performance. It does not provide one HLALA-only percentage for a complete panel.
The original 2019 HLA*LA paper reported a 99.4 percent mean across six classical HLA genes in high-coverage Illumina WGS data. That figure comes from a different benchmark and reference dataset, so it is not directly comparable with the GenDX assay comparison.
What “valid” means here
The evidence supports HLA*LA as a technically capable method for the WGS setups studied. It also shows that agreement depends on the locus, sample, coverage and comparison resolution. The 2026 comparison is a medRxiv preprint and had not been peer-reviewed at the time of posting; the authors disclose M42 funding; some also report support from PacBio, Illumina, Genentech and ONT, including travel support from ONT.
This does not establish clinical validity for every BAM file or individual sample. An HLALA call from WGS alone is not laboratory confirmation for transplant matching or other medical decisions. Genome presents HLALA results as technical output that requires expert interpretation.
Sources
- Cremin et al. (2026): Benchmarking HLA genotyping from whole genome sequencing across multiple sequencing technologies — medRxiv preprint
- Marzouka et al. (2024): Investigating the genetic makeup of the MHC in the UAE through next-generation sequencing — Scientific Reports
- Dilthey et al. (2019): HLA*LA—HLA typing from linearly projected graph alignments — Bioinformatics