Assessment of MuTect2 and VarScan2 for somatic mutation detection in exome sequencing

Next generation sequencing is generally performed to identify somatic mutations in cancer, with increasing use in, not only research, but also for diagnosis of clinical oncological patients to personalize and improve treatments. Somatic variant callers need two sets of sequencing data, one from cancer tissue and its normal tissue counterpart, to compare and detect somatic mutations. There are many somatic variant callers to choose from, but few comparison papers have been published, and therefore it is pivotal to find an efficient way of comparison between these tools, as there is no standard for detection of somatic mutations. An assessment of two somatic variant callers, MuTect2 and VarScan2, was performed on two matching data samples, tumoral and non-tumoral, acquired from the publication "SomaticCombiner: improving the performance of somatic variant calling based on evaluation tests and a consensus approach." by Wang et al. (1). We hypothesized that MuTect2 would perform better with cancer samples, as it employs the probabilistic framework of Bayesian statistics, used by most existing variant callers. Their performance was analyzed in both synthesized and real cancer samples. Both variant callers performed similarly in different samples, although VarScan2 usually surpassed MuTect2 when mutation frequency was high, MuTect2 was more consistent throughout all mutation frequencies. We found out that VarScan2 has a higher number of concordant mutations at high frequencies but, when they drop below 20%, MuTect2 performs better identifying up to 4000 mutations to VarScan2's 1000. Similarly, at frequencies over 40%, VarScan2 has a lower rate of missing mutations than MuTect2. Also, VarScan2 had a higher recall and higher precision than MuTect2. However, through the measuring of the F1-Score, MuTect2 proved to cover a wider range of accuracy for different mutation frequencies. MuTect2 outperforms VarScan2 in the synthetic data, as well as most of the data acquired from cancer patients.