(653c) A Novel Multi-Layer Inference Approach to Reconstruct Conditional-Specific Gene Network

The recent advent of high-throughput microarray data has enabled the global analysis of the transcriptome, driving the development and application of computational approaches to study gene regulation on the genome scale, by reconstructing in silico the regulatory interactions of the gene network. Our approach differs from most of the currently available approaches by inferring transcriptional and post-transcriptional regulation. In contrast to approaches that infer both types of regulation, our methodology captures biological features at different layers rather than use a black box approach to “learning” the interaction. Furthermore, our framework does not require perturbation on each gene or time-series data to uncover the regulatory interactions. Thus far, the development and application of the theory and tools for network inference have been predominantly based on statistical dependencies between gene expressions. These methods have been shown to capture only part of the relevant biological information. These current methods generate many candidates/hypotheses when applied to real biological systems, making it difficult to experimentally assess each one. Furthermore a critical analysis of the underlying assumptions of most in silico learning approaches applied to expression data [1] demonstrate that statistical dependencies alone do not readily provide direct mechanistic insights or interactions from functional genomics. We aim to address these challenges and propose a novel multi-layer network inference approach which integrates gene expression data and transcriptional regulation information to recover a specific gene network for a given condition. Our reverse engineering framework consists of three layers. The first layer integrates microarray data from a diverse set of conditions to provide a common context of gene behaviors and then identifies the most specific candidate genes for a given condition. In the second layer we apply a state-of-art causal filtering method that combines information on the transcriptional regulatory network to reconstruct the regulatory pathway to the candidate genes. In the third layer we develop a novel method of inferring transcription factor activity to identify the important regulators in the gene network that can account for post-transcriptional regulation. By integrating multiple layers of learning, our framework captures different biological features in the transcriptional regulation to achieve an accurate reconstruction of condition-specific gene networks. We establish the accuracy of our methodology against a synthetic dataset as well as the yeast dataset.  Finally we extend the framework to the application of human diseases, such as cancer.

[1] Wu, M and Chan, C. (2011) Learning transcriptional regulation on a genome scale: a theoretical analysis based on gene expression data, Briefings in Bioinformatics, accepted