Application of Feature Allocation Models on the Inference of Tumor Heterogeneity Using Integrated Genomics Data

• Activity Number: 367
• Type: Invited
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yuan Ji*+ and Peter Mueller and Juhee Lee and Kamalakar Gulukota
• Companies: NorthShore University HealthSystem and University of Texas at Austin and University of California, Santa Cruz and NorthShore University HealthSystem
• Keywords: Feature Allocation Models ; Indian Buffet Process ; Next-Generation Sequencing ; Tumor Heterogeneity ; Subclones

Abstract

A hallmark of the dynamic evolution of cancer is the presence of superceding clonal expansions, driven by shifting selective pressures, mutational processes, and disrupted oncogenes or tumor suppressors. These processes mark the genome of cancer cells in a tumor in a way such that each tumor's life history is imprinted in the somatic mutations that can be found in that tumor. New somatic mutations give rise to new cellular subpopulations called subclones. In this way, the described process of somatic mutations induces the often observed tumor heterogeneity. We propose new feature allocation models to decipher this narrative and infer the resulting subclonal diversification within and between tumors, i.e., intra- and inter-tumor heterogeneity. The proposed inference is based on feature allocation models (Broderick et al., 2013). Specifically, we use variations of the Indian buffet process (IBP) to define subclones as distinct sets of mutational incidences at these loci. In short, different subclones are the features and somatic point mutations are the experimental units that select or do not select these features. We also extend inference models to integrate multi-modal genomics fea