Dynamic Gene Coexpression Analysis with Correlation Modeling

TL;DR

This paper introduces a novel statistical method for modeling dynamic gene coexpression that varies with continuous covariates like genetic ancestry, demonstrating higher power and computational efficiency in simulations and real data analysis.

Contribution

It presents a new covariance modeling approach for dynamic gene coexpression analysis, including a simple score test and expansion to network-level investigations.

Findings

Higher statistical power than existing methods

Effective in diverse scenarios including real data

Computationally efficient for large datasets

Abstract

In many transcriptomic studies, the correlation of genes might fluctuate with quantitative factors such as genetic ancestry. We propose a method that models the covariance between two variables to vary against a continuous covariate. For the bivariate case, the proposed score test statistic is computationally simple and robust to model misspecification of the covariance term. Subsequently, the method is expanded to test relationships between one highly connected gene, such as a transcription factor, and several other genes for a more global investigation of the dynamic of the coexpression network. Simulations show that the proposed method has higher statistical power than alternatives, can be used in more diverse scenarios, and is computationally cheaper. We apply this method to African American subjects from GTEx to analyze the dynamic behavior of their gene coexpression against genetic ancestry and to identify transcription factors whose coexpression with their target genes change with the genetic ancestry. The proposed method can be applied to a wide array of problems that require covariance modeling.