Computational Inference of Genetic Regulatory Networks in Human Cancer Cells

Adam Margolin

Computer Science, Columbia University

Deconvoluting the complex network of molecular interactions that give rise to cellular phenotypes is a critical challenge in systems biology. High throughput technologies have allowed the simultaneous measurement of the concentrations of thousands of molecular species in a biological system, such as mRNA, microRNA, proteins and metabolites. As the dynamics of each molecular species is influenced by the concentration of several other species, each experiment may be treated as an observation from a joint probability distribution (JPD). The inference task is to identify statistical dependencies within this JPD to provide clues about the underlying interaction network. In the first part of this talk, I will describe an information theoretic algorithm, named ARACNE, aimed at identifying direct genetic regulatory interactions via irreducible statistical dependencies between gene expression profiles. Despite limitations that exist with all such "reverse engineering" algorithms in terms of data size requirements and limited monitoring of the relevant molecular species (i.e. only mRNAs), predictions made by this approach have been robustly validated biochemically. In the second part of this talk I will present a novel algorithm for the analysis of ChIP-on-chip data that reveals ubiquitous transcription factor (TF) / DNA interactions, and helps explain the success of reverse engineering algorithms. This algorithm predicts multiple times more TF/DNA interactions than previous algorithms, and these predictions have been preliminarily confirmed by chromatin immunoprecipitation experiments. Finally, based on this observation, I present an algorithm that uses a novel multivariate statistical approach, based on information theory, to infer genes that interact in a pathway to cooperatively regulate a large set of common targets. I describe the relationship between this algorithm and information theoretic, maximum entropy, and probabilistic graphical models frameworks, and I describe extensive biochemical validation experiments that confirm predicted post translational regulators and co-transcription factors that interact with the MYC proto-oncogene in human B cells.