For each site we denote the reference allele as r ∈ {A, C, G, T} and denote by bi and ei the called base of the i-th read (i=1…d) that covers the site and the probability of error of that base call (each base has an associated Phred-like quality score qi where ei=10-qi10). To call a variant in the tumor we try to explain the data using two models: (i) a model, M0, in which there is no variant at the site and all non-reference bases are explained by sequencing noise; and (ii) a model, Mfm, in which a variant allele m truly exists at the site with an allele fraction f and, as in M0, reads are also subject to sequencing noise. Note that M0 is equivalent to Mfm with f=0.
