To overcome these limitations, we propose a probabilistic approach based on a Binomial mixture model, called SNVMix1, which computes posterior probabilities, providing a measure of confidence on the SNV predictions. The model infers the underlying genotype at each location. We assume the genotype to be in one of three states: aa = homozygous for the reference allele, ab = heterozygous and bb = homozygous for the non-reference allele; the latter two genotypes constituting an SNV. In Figure 2, we show how the posterior probability of each of these three states increases with more depth, which demonstrates the theoretical qualities of our approach. Two other approaches: Maq (Li,H. et al., 2008) and SOAPSNP (Li,R. et al., 2008) have proposed using Binomial distributions to model genotypes; however, these were developed in the context of sequencing normal genomes, not cancer genomes. Both set parameters for the model assuming expected distributions for normal allelic ratios, and apply post-processing heuristics to reduce false positives. In our application, we are interested in cancer genomes and transcriptomes, both of which may not follow expected distributions due to
