Efficient Gaussian sample specific network marker discovery and drug enrichment analysis validation

Abstract

Identifying stable gene markers at an individual level can help to understand the genetic mechanisms of each individual patient and accomplish personalized medicine. In this paper, we propose an efficient framework to identify sample-specific markers. Gene expression data first is transformed to a corresponding likelihood matrix to alleviate inherent noise besides adding population information to each sample. Then those significantly differential genes or gene pairs are further mapped to a STRING network for analysis by assuming that the likelihood of each gene or gene pairs in the control group follows a Gaussian distribution. The proposed method is applied to three benchmark datasets including lung adenocarcinoma, kidney renal clear cell carcinoma, and uterine corpus endometrial carcinoma. It is found that disease gene markers identified by the proposed methods outperform the previous sample-specific network (SSN) method in both subtyping and survival analysis. Furthermore, we exploit the application of the subtype markers in following drug selection. The difference of the enriched drug set may reflect some underlying mechanisms of the subtypes and shed light on selecting appropriate drugs for each cancer subtype.

Keywords: Cancer; Drug; Gaussian distribution; Network biomarkers.