Extending biological pathways by utilizing conditional mutual information extracted from rna-seq gene expression data

Abstract

We propose a method of constructing a gene/protein regulatory network specifically tailored for assessing the disease state of a patient by combining generally known gene/protein pathways with transcription level changes obtained by comparing the patient’s data with the average gene expression data of the disease population. This approach uses histogram estimation with conditional mutual information to identify if some genes/proteins may more likely interact with each other. We applied our method to the Cancer Genome Atlas (TCGA) cancer data, specifically, RNA-Seq gene expression data of 110 breast cancer, 141 colorectal cancer, 445 gastric cancer and 105 rectal cancer samples, which are publicly available. We focused on examining transcription factors such as SNAI1, SNAI2, ZEB2, and TWIST1 and their downstream targets in EMT pathway (e.g., OCLN, DSP, VIM and CDH2…). We discovered that although the participating biological entities of the EMT pathway are generally known, our approach can extend their regulatory relationships through new discoveries. Our approach could form a basis for inventing a novel way of constructing a gene regulation pathway specifically tailored for each individual cancer patient.