Detection of gene-gene interactions using multistage sparse and low-rank regression

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• High-dimensional data presents computational challenges in biological science, especially when studying interactions.
• Existing methods struggle with the curse of dimensionality, limiting the accurate detection of gene-gene interactions.

Purpose of the Study:

• To propose a novel sparse and low-rank (SLR) screening method for efficient and accurate detection of gene-gene interactions.
• To enhance statistical analysis performance in high-dimensional biological datasets.

Main Methods:

• Developed a sparse and low-rank (SLR) screening approach combining a low-rank interaction model with Lasso screening.
• Modeled interaction effects using a low-rank matrix for parsimonious parametrization.
• Integrated SLR screening into the Screen-and-Clean framework to mitigate issues like Bonferroni correction penalties.

Main Results:

• The SLR screening method demonstrated improved accuracy in detecting gene-gene interactions compared to conventional methods.
• The procedure successfully identified main and interaction effects that were previously overlooked.
• Application to the Warfarin dosage and CoLaus studies validated the effectiveness of the SLR approach.

Conclusions:

• The proposed SLR screening offers a computationally feasible and efficient solution for analyzing high-dimensional biological data.
• SLR screening enhances the ability to identify complex gene-gene interactions, advancing biological discovery.
• This method provides a robust tool for statistical inference in genomics and related fields.