A comparative study of methods for detecting small somatic variants in disease-normal paired next generation sequencing data

While high-throughput next generation sequencing technologies are rapidly approaching maturity, computational tools for variant calling have significant room for improvement. The recently emerged computational methods typically compare a disease sample directly with its matched control in order to improve the accuracy of variant calling and search disease specific mutations. In this paper, we performed a comparative study of five methods, JointSNVMix, SAMtools, SomaticSniper, Strelka and VarScan, for simultaneous detection of small somatic variants from disease-normal pairs. This paper evaluates the sensitivity and false discovery rate of these methods, aiming to provide guidelines for users and algorithm developers. The computational efficiency and other issues are also explored.