Dissecting Cancer Heterogeneity with a Probabilistic Genotype-Phenotype Model

Developing an approach to model heterogeneity of cancer has emerged as is an urgent need in cancer studies. To address this challenge we propose an approach for a probabilistic modeling of cancer. Starting with the assumption that each cancer case should be consider as a mixture of cancer subtypes, our model links phenotypic similarities with putative causes. Specifically, building on the idea of a topic model [1], our approach is based on two components (i) a measure of phenotypic similarity between the patients and (ii) a list of features -such as mutations, copy number variation, microRNA level etc. to be used as proposed explanations. The main idea is to define (probabilistic) disease subtypes and, for each patient, identify the mixture of the subtypes that best explain the patient similarity network. Our approach does not assume predefined subtypes nor does it assume that such subtypes have to be uniquely defined. That is, we do not assume that there exist “the” disease subtype model but rather we consider a distribution of such models providing a probabilistic context.