Network regression analysis in transcriptome-wide association studies.

Area of Science:

• Genetics
• Bioinformatics
• Systems Biology

Background:

• Transcriptome-wide association studies (TWASs) integrate genome-wide association studies (GWASs) and eQTL data to uncover disease mechanisms.
• Existing TWAS methods often analyze genes individually, neglecting crucial gene interdependencies and network structures vital for complex diseases.
• This limitation can reduce the statistical power to detect associations involving multiple interacting genes.

Purpose of the Study:

• To develop a novel method, NeRiT, for detecting associations between biological networks and traits of interest.
• To address the limitations of single-gene TWAS approaches by incorporating gene network information.
• To improve the power and accuracy of TWAS by accounting for gene interdependencies.

Main Methods:

• Developed Network Regression in a two-stage TWAS framework (NeRiT).
• Utilized Bayesian Dirichlet process regression for gene expression prediction weights in the first stage.
• Employed pointwise mutual information to model gene-node correlations and network structure in the second stage.

Main Results:

• NeRiT demonstrated calibrated type I error control and superior power compared to existing methods, particularly for detecting edge effects.
• The method's performance remained robust across various simulation parameters, including GWAS sample size and network structures.
• Real-world application on UK Biobank data successfully identified trait-related nodes and edges for blood pressure traits.

Conclusions:

• NeRiT is a powerful and efficient network regression method for TWAS.
• The approach effectively leverages gene network information to enhance the discovery of trait-associated genes and their interactions.
• NeRiT offers a significant advancement in understanding complex disease genetics through integrated network analysis.