Kernel-based hierarchical structural component models for pathway analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Pathway analysis is crucial for interpreting omics data but often overlooks pathway correlations and assumes linear biomarker-phenotype associations.
• Existing methods may yield misleading results due to ignoring pathway overlap and complex biological relationships.

Purpose of the Study:

• To develop a novel pathway analysis approach that models complex, non-linear biomarker-phenotype associations.
• To simultaneously analyze entire biological pathways while considering the hierarchical structure of biomarkers.

Main Methods:

• Proposed Hierarchical structural CoMponent analysis using Kernel (HisCoM-Kernel), extending kernel machine regression.
• Incorporated biomarker-pathway hierarchical structure for simultaneous pathway analysis.
• Applied the method to diverse omics datasets (RNA-seq, etc.).

Main Results:

• HisCoM-Kernel demonstrated higher statistical power in simulation studies compared to existing pathway-based methods.
• The method successfully identified biologically meaningful pathways across three different omics datasets.
• Showcased superior performance in identifying relevant pathways, including those previously reported.

Conclusions:

• HisCoM-Kernel offers a flexible and powerful approach for pathway analysis in various omics data.
• The method effectively models complex biological interactions, improving the interpretability of omics studies.
• HisCoM-Kernel enhances the discovery of biologically relevant pathways by addressing limitations of current approaches.