A Bayesian Nonparametric Model for Integrative Clustering of Omics Data

Cancer is a complex disease, driven by a range of genetic and environmental factors. Many integrative clustering methods aim to provide insight into the mechanisms underlying cancer but fewof them are computationally efficient and able to estimate the number of subtypes. We have developed a Bayesian nonparametric model for combined data integration and clustering called BayesCluster, which aims to identify cancer subtypes and addresses many of the issues faced by the existing integrative methods. The proposed method can integrate and use the information from multiple different datasets, and offers better cluster interpretability by using nonlocal priors. We incorporate feature learning because of the large number of predictors, and use a Dirichlet process mixture model approach to produce the patient subgroups. We ensure tractable inference with simulated annealing. We apply the model to datasets from the Cancer Genome Atlas project of glioblastoma multiforme, which contains clinical and biological data about cancer patients with extremely poor prognosis of survival. By combining all available information we are able to be better identify clinically meaningful subtypes of glioblastoma.