Tumor cell populations can be thought of as being composed of homogeneouscell subpopulations, with each subpopulation being characterized by overlappingsets of single nucleotide variants (SNVs). Such subpopulations are known assubclones and are an important target for precision medicine. Reconstructingsuch subclones from next-generation sequencing (NGS) data is one of the majorchallenges in precision medicine. We present PairClone as a new tool toimplement this reconstruction. The main idea of PairClone is to model shortreads mapped to pairs of proximal SNVs. In contrast, most existing methods useonly marginal reads for unpaired SNVs. Using Bayesian nonparametric models, weestimate posterior probabilities of the number, genotypes and populationfrequencies of subclones in one or more tumor sample. We use the categoricalIndian buffet process (cIBP) as a prior probability model for subclones thatare represented as vectors of categorical matrices that record thecorresponding sets of mutation pairs. Performance of PairClone is assessedusing simulated and real datasets. An open source software package can beobtained at http://www.compgenome.org/pairclone.
展开▼
