Abstract
Short tandem repeat (STR) variation is an often overlooked source of variation between genomes. STRs comprise about 3% of the human genome and are highly polymorphic. Some cause Mendelian disease, and others affect gene expression. Their contribution to common disease is not well-understood, but recent software tools designed to genotype STRs using short read sequencing data will help address this. Here, we compare software that genotypes common STRs and rarer STR expansions genome-wide, with the aim of applying them to population-scale genomes. By using the Genome-In-A-Bottle (GIAB) consortium and 1000 Genomes Project short-read sequencing data, we compare performance in terms of sequence length, depth, computing resources needed, genotyping accuracy and number of STRs genotyped. To ensure broad applicability of our findings, we also measure genotyping performance against a set of genomes from clinical samples with known STR expansions, and a set of STRs commonly used for forensic identification. We find that HipSTR, ExpansionHunter and GangSTR perform well in genotyping common STRs, including the CODIS 13 core STRs used for forensic analysis. GangSTR and ExpansionHunter outperform HipSTR for genotyping call rate and memory usage. ExpansionHunter denovo (EHdn), STRling and GangSTR outperformed STRetch for detecting expanded STRs, and EHdn and STRling used considerably less processor time compared to GangSTR. Analysis on shared genomic sequence data provided by the GIAB consortium allows future performance comparisons of new software approaches on a common set of data, facilitating comparisons and allowing researchers to choose the best software that fulfils their needs.