Mitochondrial DNA (mtDNA) variants have been implicated in a wide range of diseases. With the advancement and reduced cost of whole-genome sequencing (WGS), mtDNA variant analysis has become increasingly feasible for both clinical diagnostics and large-scale population studies. However, determining which bioinformatic tools are most suitable for accurate mtDNA variant detection from short-read WGS remains an open question. Several recently developed tools aim to address the unique challenges of mtDNA variant calling, such as misalignment caused by nuclear mitochondrial DNA segments (NUMTs) and the circular nature of the mitochondrial genome, yet no comprehensive benchmarking has been available.
Here, we systematically compared four tools that can be applied to large-scale WGS datasets on high-performance computing (HPC) and cloud platforms: mitoHPC, mtSwirl, mity, and mtDNA Server 2. Using both empirical data from the 1000 Genomes Project (1000G) and simulated mixtures of samples from distinct haplogroups, we evaluated performance across homoplasmic and heteroplasmic variant detection. Our results show that although mity and mtDNA Server 2 are fast and user-friendly, mtSwirl and mitoHPC achieved higher accuracy for both variant types. Additionally, we generated a mitochondrial genome principal component analysis (PCA) reference using the 1000G WGS data, providing a resource for researchers to compare their mtDNA PCA results and facilitate mitochondrial genome-wide association studies (mtGWAS).