Gene Regulatory Network Reconstruction and Pathway Inference from High Throughput Gene Expression Data.

Abstract

Two basic motivating questions in biomedical research are: What genes regulate what other genes? What genes or groups of genes regulate a specific phenotype? Gene regulatory network (GRN) reconstruction and pathway inference are the two computational strategies addressing these two questions respectively. GRN reconstruction is to infer the components and topology of an unknown pathway, while pathway inference is to infer association between known pathways and a phenotype. This thesis focuses on gene regulatory network reconstruction and pathway inference from high throughput biological data. In the first part of this work, I developed a novel method, MI3, for de novo GRN reconstruction using continuous three-way mutual information. MI3 addresses three major issues in previous probabilistic methods simultaneously: (1) to handle continuous variables, (2) to detect high order relationships, (3) to differentiate causal vs. confounding relationships. MI3 consistently and significantly outperformed frequently used control methods and faithfully capture mechanistic relationships from gene expression data. In the second part of this work, I proposed another novel method, GAGE, Generally Applicable Gene Set Enrichment for pathway inference. I successfully apply GAGE to multiple microarray data sets with different sample sizes, experimental designs and profiling techniques. GAGE shows significantly better performance when compared to two other commonly used GSA methods of GSEA and PAGE. GAGE reveals novel and relevant regulatory mechanisms from both published and previously unpublished microarray studies. In the third part of this work, we conducted a microarray study on transcriptional programs during BMP6 induced osteoblast differentiation and mineralization, and applied GAGE to recover the regulatory pathways and transcriptional signaling networks in the process. I not only showed which pathways or gene sets are significant, but also when and how they are involved in the osteoblast differentiation and mineralization. Different from common pathway analyses, our work further captures the interconnections among individual pathways or functional groups and integrate them into a whole system.