High-throughput sequencing technologies have facilitated the generation of an unprecedented amount of genomic cancer data, opening the way to a more profound understanding of tumorigenesis. In this endeavor, two fundamental questions have emerged, namely (1) which alterations drive tumor progression and (2) in which order do they occur? Answering these questions is crucial for therapeutic decisions involving targeted agents. Because of interpatient heterogeneity, progression at the level of pathways is more reproducible than progression at the level of single genes. In this study, we introduce pathTiMEx, a generative probabilistic graphical model that describes tumor progression as a partially ordered set of mutually exclusive driver pathways. pathTiMEx employs a stochastic optimization procedure to jointly optimize the assignment of genes to pathways and the evolutionary order constraints among pathways. On real cancer data, pathTiMEx recapitulates previous knowledge on tumorigenesis, such as the temporal order among pathways which include APC, KRAS, and TP53 in colorectal cancer, while also proposing new biological hypotheses, such as the existence of a single early causal event consisting of the amplification of CDK4 and the deletion of CDKN2A in glioblastoma. pathTiMEx is available as an R package.
Keywords: cancer genomics; cancer pathways; cancer progression; mutual exclusivity; probabilistic graphical models.