CoDP: Cooperative Driver Pathways Discovery With Matrix Factorization and Tri-Random Walk

Abstract

Discovering driver pathways is becoming a key step to uncover the molecular mechanism underlying cancer and to explore precise treatments for cancer patients. Most of the previous works focus on identifying individual pathways. In practice, multiple pathways often cooperatively trigger cancer. In this study, we propose a new approach called CoDP to discover cooperative driver pathways. CoDP firstly uses integrative nonnegative matrix factorization to identify cancer-related comodules (composed with genes and miRNAs) by fusing four types of data, including the expression profiles of miRNAs and genes, miRNA-gene interaction data, and gene-gene interaction data of cancer samples. Next, CoDP builds a hybrid network with nodes corresponding to pathways, to genes and miRNAs in comodules, and it further uses tri-random walk on the subnetworks of the hybrid network to update and replenish associations between genes and pathways, and those between miRNAs and pathways. After that, CoDP introduces a quantitative function to quantify the coverage of pathway sets for each comodule based on replenished associations, and identifies the pathway set with the highest coverage as cooperative driver pathways. Empirical study on ovarian (liver) cancer datasets shows that in the identified comodules, miRNAs cooperatively regulate genes, genes (i.e., TP53) and miRNAs (i.e., mir-10b) are statistical significantly associated with the cancer, and others may have potential associations with the cancer. The identified cooperative pathways are involved with key biological processes and carcinogenesis of ovarian (liver) cancer. CoDP also uncovers more known driver genes (over 300%) and cooperative driver pathways (over 200%) than existing methods. The codes of CoDP are available at http://mlda.swu.edu.cn/codes.php?name=CoDP.