Discovering genomic associations on cancer datasets by applying sparse regression methods

Abstract

Association analysis of gene expression traits with genomic features is crucial to identify the molecular mechanisms underlying cancer. In this study, we employ sparse regression methods of Lasso and GFLasso to discover genomic associations. Lasso penalizes a least squares regression by the sum of the absolute values of the coefficients, which in turn leads to sparse solutions. GFLasso, an extension of Lasso, fuses regression coefficients across correlated outcome variables, which is especially suitable for the analysis of gene expression traits having inherent network structure as output traits. Our study is about considering combined benefits of these computational methods and investigating the identified genomic associations. Real genomic datasets from breast cancer and ovarian cancer patients are analyzed by the proposed approach. We show that the combined effect of both the methods has a significant impact in identifying the crucial cancer causing genomic features with both weaker and stronger associations.