Extended Data Fig. 7: Stratifying predicted MHC/KIR alleles.

On each subplot, four bars represent Locityper with the full reference panel (F); Locityper in the leave-one-out setting without and with weights (denoted L and W, respectively); and T1K. Genotyping is performed for 40 HPRC samples across 40 (pseudo)genes from the MHC locus (a) and 17 (pseudo)genes from the KIR locus (b). T1K/Locityper allele predictions are placed into six categories: extra copy for cases when a genotyper called more gene copies than actually present in the locus; missing copy when a genotyper failed to call a present gene copy; (in)correct protein for predictions where a protein product (second field in the HLA/KIR nomenclature) was called (in)correctly; unavailable protein for such Locityper LOO predictions, where true protein product is unavailable in the LOO database and therefore cannot be correctly identified; and deletion found for cases when a genotyper correctly identified a missing gene copy. Last entry in each panel shows average fraction across all MHC/KIR genes/pseudogenes.