Benchmarking the HLA typing performance of three HLA assays and seven NGS-based HLA algorithms

Abstract Background With the great progress made recently in NGS (Next Generation Sequencing) technology, sequencing accuracy and throughput have increased, while the cost for data has decreased. Various HLA (Human Leukocyte Antigen) typing algorithms and assays have been developed and have begun to be used in clinical practice. However, there is no systematic benchmarking to evaluate the HLA typing performance of different HLA assays and algorithms. In this study, we compared the HLA typing performance of three HLA assays and seven NGS-based HLA algorithms and assessed the impact of sequencing depth and length on HLA typing accuracy. Results Seven HLA typing algorithms at 4- and 6-digit allele levels were compared on three different assays in terms of accuracy, read depth and read length. The algorithms HLA-HD and HISAT-genotype showed the highest accuracy at both 2- and 4-digit resolution, followed by HLAscan. We designed a capture-based HLA assay, which showed comparable or even better performance compared with WES (Whole Exome Sequencing). In the depth evaluation, the sequencing data were down-sampled from 500X to 10X based on the depth of HLA genes. We found that the minimal depth was 100X for HLA-HD and HISAT-genotype to obtain more than 90% HLA typing accuracy at the 6-digit allele level. The accuracy of all three algorithms did not change when the read length decreased from 150 bp to 76 bp. Conclusion Although HISAT-genotype and HLA-HD may need more computing resources, we recommend using them for NGS-based HLA genotyping because of their higher accuracy and robustness to sequencing depth and read length. We propose that the minimal sequence depth for obtaining more than 90% HLA typing accuracy at the 6-digit allele level is 100X. Besides, targeting capture-based NGS HLA typing may be more suitable than WES in clinical practice due to its lower sequencing cost and higher HLA sequencing depth.