Accurate imputation of human leukocyte antigens with CookHLA

CookHLA substantially improved imputation accuracy over previous methods, including the predecessor SNP2HLA by implementing several changes.
First, to increase accuracy and efficiency, we employed the recently released hidden Markov model (Beagle v4 and v5). Second, to further improve accuracy, we developed a procedure that can account for local variability in the exons of HLA genes. Highly polymorphic exons (exon 2/3/4 in Class I and 2/3 in Class II) are critical regions whose sequences can determine most of the 4-digit alleles. In our method, we repeat imputation by putting the marker set locally in each of the highly polymorphic exons, and use consensus posterior probabilities from the repeated analyses for final predictions (Figure B). Third, to increase accuracy even further, we adaptively learn the genetic map of MHC from the data. This map information allows us to account for the data-specific LD structure within MHC, which improves the imputation accuracy compared to the use of publicly available genetic maps (Figure C).