Introduction
Accurate detection of somatic point mutations remains a challenge, due to the admixture of normal cells and the presence of multiple subclones of tumor cells. We have developed a novel Bayesian phylogenetic method, MuSE (mutation somatic evolution estimation), for describing the evolution from the reference allele to the tumor and the normal allelic composition at a single nucleotide position. Our proposed method incorporates the probability of sequencing errors and computes the unknown allele frequencies, multiple alternative alleles and the rates of nucleotide transition/transversion. All model parameters are estimated using the maximum likelihood or the Markov chain Monte Carlo (MCMC) method. By comparing the somatic variant allele fraction (π) between the paired tumor–normal samples, we classify variants into the following categories: somatic, germ-line and reversal to the homozygous reference. We also include filters that consider the sequence context surrounding the point mutations, in order to further reduce the number of false positives.
We have validated the performance of MuSE using a virtual-tumor benchmarking approach as previously described (Cisbulskis et al.), and applied it to analysis of TCGA exome sequencing data, such as Kidney Chromophobe (KICH), Kidney renal papillary cell carcinoma (KIRP) and Adrenocortical carcinoma (ACC).
Version
Available upon request.
Follow MuSE performance in the DREAM mutation calling challenge
Check the leaderboard for the winners of each round. MuSE won the 2nd place in calling single nucleotide variants.