Gene-gene interaction analysis incorporating network information via a structured Bayesian approach

TL;DR

This paper introduces a structured Bayesian approach for gene-gene interaction analysis that incorporates network information, improving the identification of biologically meaningful interactions and phenotype prediction accuracy.

Contribution

It is among the first to integrate network selection and structures into gene-gene interaction analysis using a Bayesian framework with variational inference.

Findings

Outperforms existing methods in simulation studies

Identifies biologically relevant gene interactions in TCGA data

Achieves high prediction accuracy and stable gene selection

Abstract

Increasing evidence has shown that gene-gene interactions have important effects on biological processes of human diseases. Due to the high dimensionality of genetic measurements, existing interaction analysis methods usually suffer from a lack of sufficient information and are still unsatisfactory. Biological networks have been massively accumulated, allowing researchers to identify biomarkers from a system perspective by utilizing network selection (consisting of functionally related biomarkers) as well as network structures. In the main-effect analysis, network information has been widely incorporated, leading to biologically more meaningful and more accurate estimates. However, there is still a big gap in the context of interaction analysis. In this study, we develop a novel structured Bayesian interaction analysis approach, effectively incorporating the network information. This study is among the first to identify gene-gene interactions with the assistance of network selection for phenotype prediction, while simultaneously accommodating the underlying network structures. It innovatively respects the multiple hierarchies among main effects, interactions, and networks. Bayesian method is adopted, which has been shown to have multiple advantages over some other techniques. An efficient variational inference algorithm is developed to explore the posterior distribution. Extensive simulation studies demonstrate the practical superiority of the proposed approach. The analysis of TCGA data on melanoma and lung cancer leads to biologically sensible findings with satisfactory prediction accuracy and selection stability.