With the dramatic expansion of cancer sequencing, research in detection algorithms has been increasingly concentrating on somatic mutations. Many software packages have been proposed, e.g. SNVMix, SomaticSniper, SoapSNP, and Strelka, but none yet furnish an adequate solution to the problem, primarily because of the difficulty in properly handling cancer-specific anomalies like purity issues and clonality. We are developing a tool called Bassovac that, unlike prior tools, does not use ad-hoc modeling of these effects, but rather treats them fully probabilistically. Specifically, all effects are integrated at the atomic level of the "read" and standard probability theory integrates read tallies to the sample level and then further to the tumor-normal pair level. Bayes Theorem is then used for inversion to obtain the posterior probability of somatic mutation.
Bassovac is pending public release. Preliminary results on simulated data and WGS data from several cancer types (prostate, AML, breast) indicate significantly higher sensitivity than other algorithms, particularly for sub-clonal events having <10% tumor variant allele frequency.