Information-theoretical analysis of gene expression data to infer transcriptional interactions

Abstract

The majority of human diseases are related with the dynamic interaction of many genes and their products as well as environmental constraints. Cancer (and breast cancer in particular) is a paradigmatic example of such complex behavior. Since gene regulation is a non-equilibrium process, the inference and analysis of such phenomena could be done following the tenets of non-equilibrium physics. The traditional \emph{programme} in statistical mechanics consists in inferring the joint probability distribution for either microscopic states (equilibrium) or mesoscopic-states (non-equilibrium), given a model for the particle interactions (e.g. the potentials). An \emph{inverse problem} in statistical mechanics, in the other hand, is based on considering a \emph{realization} of the probability distribution of micro- or meso-states and used it to infer the interaction potentials between particles. This is the approach taken in what follows. We analyzed 261 whole-genome gene expression experiments in breast cancer patients, and by means of an information-theoretical analysis, we deconvolute the associated set of transcriptional interactions, i.e. we discover a set of fundamental biochemical reactions related to this pathology. By doing this, we showed how to apply the tools of non-linear statistical physics to generate hypothesis to be tested on clinical and biochemical settings in relation to cancer phenomenology.