Testing Differential Networks with Applications to Detecting Gene-by-Gene Interactions

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genomics

Background:

• Gene-gene interactions significantly influence complex traits' genetic variation.
• High dimensionality of gene interaction data poses methodological challenges for identification.
• Existing methods struggle with the complexity of gene networks in relation to sample size.

Purpose of the Study:

• To develop a robust statistical framework for identifying gene-by-gene interactions associated with binary traits.
• To translate the complex problem of interaction detection into a manageable precision matrix comparison.
• To provide powerful and statistically sound methods for analyzing differential gene networks.

Main Methods:

• Utilized a Gaussian graphical model framework to represent gene dependencies.
• Framed gene-gene interaction identification as comparing two high-dimensional precision matrices.
• Developed a global network difference test powerful against sparse alternatives.
• Implemented a false discovery rate-controlled multiple testing procedure for specific network structure inference.

Main Results:

• Proposed methods effectively test for differential gene networks, particularly under sparsity.
• Theoretical guarantees ensure test validity and optimality under specified assumptions.
• Simulation studies confirmed desired error rates and good statistical power.
• The approach was successfully applied to a real-world breast cancer gene expression dataset.

Conclusions:

• The Gaussian graphical model approach provides a powerful tool for identifying gene-gene interactions in complex traits.
• The developed methods offer reliable inference of differential gene networks with controlled error rates.
• This framework enhances our understanding of genetic architecture in diseases like breast cancer.