A multi-task convolutional deep learning method for HLA allelic imputation and its application to trans-ethnic MHC fine-mapping of type 1 diabetes

<jats:title>Abstract</jats:title><jats:p>Conventional HLA imputation methods drop their performance for infrequent alleles, which reduces reliability of trans-ethnic MHC fine-mapping due to inter-ethnic heterogeneity in allele frequency spectra. We developed DEEP<jats:sup>*</jats:sup>HLA, a deep learning method for imputing HLA genotypes. Through validation using the Japanese and European HLA reference panels (<jats:italic>n</jats:italic> = 1,118 and 5,112), DEEP<jats:sup>*</jats:sup>HLA achieved the highest accuracies in both datasets (0.987 and 0.976) especially for low-frequency and rare alleles. DEEP<jats:sup>*</jats:sup>HLA was less dependent of distance-dependent linkage disequilibrium decay of the target alleles and might capture the complicated region-wide information. We applied DEEP<jats:sup>*</jats:sup>HLA to type 1 diabetes GWAS data of BioBank Japan (<jats:italic>n</jats:italic> = 62,387) and UK Biobank (<jats:italic>n</jats:italic> = 356,855), and successfully disentangled independently associated class I and II HLA variants with shared risk between diverse populations (the top signal at HLA-DRβ1 amino acid position 71; <jats:italic>P</jats:italic> = 6.2 ×10<jats:sup>−119</jats:sup>). Our study illustrates a value of deep learning in genotype imputation and trans-ethnic MHC fine-mapping.</jats:p>