Computational predictions of HLA-I from shotgun data by targeted assembly (left) or read alignment (right). For targeted assembly, NGS reads having their first fifteen 5' bases matching one of HLA CDS (RNA-Seq) or genomic (WGS/exon capture) sequences are recruited and assembled de novo with TASR. Resulting sequence contigs are aligned against a database sequence of all predicted HLA CDS (RNA-Seq) or genomic sequences (WGS/exon capture), tracking best HLA hit(s). Reciprocal best alignments are considered in the same manner. Putative allele assignments from shotgun datasets (HLAminer) are informed by contig length, depth of coverage and similarity to reference sequences, when applicable. The probability of each prediction being correct is estimated by determining the probability of that prediction being observed by chance.