In addition, the Bayesian framework of QuantiSNP provides considerable flexibility for extending the model to specific applications. In cancer studies, heterogeneous samples are a common problem in which tumour samples maybe contaminated by the presence of normal gDNA. In such instances, the observed log R ratios and B allele frequencies will be a mixture of the signals due to the two sample components: 15 where (x, y) are the intensities due to each allele and μ is the mixing proportion. It is then necessary to deconvolve the mixture by estimating the mixing proportion, which may be assumed to be constant for the whole sample, from the observed data. A strength of our method is that not only is this type of inference possible, via an extension of the observation model for the HMM, it is also possible to generate artificial heterogeneous datasets with pre-specified mixing proportions (such as in (20)) in order to estimate our false positive characteristics for different mixtures. A further feature relevant for cancer studies would be the joint analysis, which should increase the ability to identify common genomic alterations in a set of cancer samples.
