TreeKernel: interpretable kernel machine tests for interactions between -omics and clinical predictors with...

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Investigating interactions between high-dimensional omics data and clinical covariates is crucial for understanding complex diseases.
• Omics data, such as metabolic pathways, can exhibit varied relationships with clinical phenotypes depending on patient covariates like age and sex.
• Existing methods may not fully capture these nuanced, covariate-dependent omics-phenotype relationships.

Purpose of the Study:

• To develop and evaluate a novel statistical method for testing associations between high-dimensional omics pathways and clinical phenotypes.
• To specifically address and model the influence of clinical covariates on the omics-phenotype relationship.
• To apply the method to identify clinically meaningful interactions in real-world datasets.

Main Methods:

• Proposing a method that partitions the clinical covariate space.
• Performing kernel association tests within these partitions to detect covariate-specific omics-phenotype relationships.
• Utilizing hierarchical partitioning for structured analysis of clinical covariates.

Main Results:

• The proposed method, TreeKernel, demonstrates superior performance over competing methods in simulation studies.
• It achieves higher statistical power in identifying differential relationships across clinical groups while controlling the Type I error rate.
• Application to the COPDGene study revealed significant interactions between metabolic pathways, clinical factors, and lung function.

Conclusions:

• TreeKernel offers a straightforward and interpretable approach for analyzing omics data and clinical outcomes.
• The method is effective in detecting interactions within clinical cohorts, enhancing the understanding of disease mechanisms.
• Its broad applicability makes it a valuable tool for diverse biomedical research studies.