UCLA

We propose Bayesian models tailored to infer complex patterns of dependence among heterogeneous sets of data.

We consider highly structured information and illustrate modeling of flexible multivariate distributions using the formalism of graphical models.

Motivating applications guiding our methodological developments come

from the field of integrative biology. In particular, we tackle two

fundamental problems: the detection of causal SNPs in pharmacogenetics studies and the assessment of differential patterns of interactions characterizing the activity of biomolecular pathways. We discuss inference based on Markov Chain Monte Carlo simulation and apply our methods to

several synthetic data sets, as well as case study data from cancer genomics.

In these settings, we show how the flexibility of the Bayesian

framework is especially attractive, since it allows for the integration of scientific information by means of prior distributions, while also soundly characterizing the problem of multiple comparisons as a decision problem.