B1101
Title: Bayesian analysis of multiple related molecular networks
Authors: Gwenael Leday - University of Cambridge (United Kingdom) [presenting]
Ilaria Speranza - Politecnico di Milano (Italy)
Leonardo Bottolo - University of Cambridge (United Kingdom)
Sylvia Richardson - MRC Biostatistics - Cambridge (United Kingdom)
Abstract: The problem of inferring and comparing multiple graphical structures from high-dimensional molecular data is considered. We propose a hierarchical Bayesian model that allows the borrowing of strength across groups of samples and the joint estimation of multiple (inverse) covariance matrices. Closed-form Bayes factors are then used to identify, say, common or group-specific structures via multiple testing. The proposed approach has the advantage of allowing directionality and the testing of biologically relevant hypotheses, such as edge losses and gains in a two-group comparison. It is also computationally very efficient, addressing problems with thousands of variables in a few seconds. We illustrate the proposed method on simulated data and various real data examples.
