Causal inference of gene regulation based on sub-network assembly

Abstract

Identifying disease-causing genes requires studying the nature of genes in the context of regulatory systems. We develop a four-step strategy to integrate both gene expression and SNP variation to make causal inference on a gene regulatory network.??? The first step towards our goal is to select gene modules on an interaction map that includes the protein-protein interaction, gene-gene interaction and well recognized pathways. Those modules are overlapping to each other and they individually are highly connected according to the gene expression profiles. The modules are also selected according to their discriminative power of distinguishing patients' disease status. The second step is to select the sequence variation underlying each of the modules. A multivariate modeling strategy is adopted to select SNPs that are strongly associated with the expression. It is close to the expression QTLs while the response we concern now is the joint variation of the whole module. The third step is to build directed sub-networks with each module and its associated SNPs. The sub-networks are then assembled with a ranking strategy that transfers the local information into global information.??? ??? The proposed procedure was applied on a cohort of oral squamous cell carcinoma patients that were profiled with both exon expression and SNP variation. The modules were selected according to their association with lymph node metastasis. We evaluated our causal inference on the focal adhesion pathway by comparing the direction of regulation to the curated information in KEGG. The causal network of our inference also showed a change of regulation direction among a few key genes in the PPAR signaling pathway.