Integrating external biological knowledge in the construction of regulatory networks from LINCS data

Abstract

The inference of gene regulatory networks is of great interest and has various applications. The recent advances in high-throughout biological data collection have facilitated the construction and understanding of gene regulatory networks in many model organisms. However, the inference of gene networks from large-scale human genomic data could be challenging. Generally, it is difficult to identify the correct regulators for each gene in the large search space, given that the high dimensional gene expression data only provides small number of observations for each gene. In this thesis, we propose a Bayesian approach integrating external data sources with knockdown data from human cell lines to infer regulatory gene networks. In particular, we assemble multiple data sources including gene expression data, genome-wide binding data, gene ontology and known pathways and employ a supervised learning framework to compute prior probabilities of regulatory relationships. We show that our integrated method improves the accuracy of inferred gene networks. We present our assessment results against benchmark method and data in different forms, figures, graphs and tables. We illustrate our results in two different human cell lines, and demonstrate the generality of our results.