Although long-read single-cell sequencing technologies, like Oxford Nanopore Technologies‘ (ONT) nanopore sequencing, continue to mature, they are not yet truly challenging short-read NGS in the sequence-based HLA genotyping space. Despite their read length advantages, the read quality often does not suffice for conventional, mapping-based approaches to genotyping. Here, we show that substantial improvements in long-read genotyping quality and efficacy can be achieved by a layered approach to genotyping, where we leverage alignment-free k-mer-based probabilistic classification and statistical learning in addition to exact sequence alignment.