MAD Bayes for Tumor Heterogeneity Feature Allocation with Non-Normal Sampling

Abstract

We propose small-variance asymptotic approximations for the inference of tumor heterogeneity (TH) using next-generation sequencing data. Understanding TH is an important and open research problem in biology. The lack of appropriate statistical inference is a critical gap in existing methods that the proposed approach aims to fill. We build on a hierarchical model with an exponential family likelihood and a feature allocation prior. The proposed approach generalizes similar small-variance approximations proposed by Kulis and Jordan (2012) and Broderick et.al (2012) for inference with Dirichlet process mixture and Indian buffet prior models under normal sampling. We show that the new algorithm can successfully recover latent structures of different subclones and is also magnitude faster than available Markov chain Monte Carlo samplers, the latter often practically infeasible for high-dimensional genomics data. The proposed approach is scalable, simple to implement and benefits from the flexibility of Bayesian nonparametric models. More importantly, it provides a useful tool for the biological community for estimating cell subtypes in tumor samples.