A Bayesian Method for Partitioning Biological Pathways Based on Evolutionary History

• Activity Number: 338
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yang Li*+ and Jun Liu and Sarah E. Calvo and Roee Gutman and Vamsi K. Mootha
• Companies: Harvard and Harvard and Broad Institute and Brown University and Harvard
• Keywords: Bayesian hierarchical modeling ; Gene function ; Evolutionary history ; Dirichlet process ; Clustering ; Genomics

Abstract

Availability of diverse genome sequences makes it possible to predict gene function based on shared evolutionary history. It was roughly known that functionally related genes tend to be gained and lost simultaneously during evolution, which motivated people to measure the co-evolution of genes for predicting gene functions. The previous methods, however, worked only on pairs of genes and ignored the phylogeny of species, which often led to biased estimation of co-evolutions. We propose a Bayesian method that is based on a mixture of hidden Markov models (HMMs) with Dirichlet process prior to cluster input genes into modules with shared evolutionary history. By modeling the evolution process with HMMs on a species tree, our inference of co-evolutions takes appropriate account of the phylogeny of species. An efficient Gibbs sampler is developed for posterior distribution computation. We applied it systematically to over 1000 human biological pathways, as well as the entire genomes of organisms. Our algorithm recovers many known evolutionary modules, and the results also reveal unanticipated evolutionary modularity as well as novel, co-evolving components within well-studied pathways.