PaIRKAT: A pathway integrated regression-based kernel association test with applications to metabolomics and COPD...

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• High-throughput omics data (genomics, metabolomics, etc.) are crucial for biological research.
• Biological pathways, represented as graphs, often contain disjoint components that reduce statistical testing power.
• Existing kernel machine methods can be limited in effectively utilizing pathway information.

Purpose of the Study:

• To develop a novel statistical method, PaIRKAT, for analyzing associations between biological pathways and phenotypes.
• To improve the power of association tests by incorporating pathway topology and addressing disconnected pathway components.
• To provide a robust and generalizable method applicable to various graph-structured biological data.

Main Methods:

• Pathway Integrated Regression-based Kernel Association Test (PaIRKAT) is proposed, a kernel machine regression method.
• Incorporates known pathway information using a graph kernel regularization approach to smooth disconnected components.
• Extends semi-parametric kernel regression frameworks with a score test utilizing a regularized graph.

Main Results:

• PaIRKAT demonstrates robustness against incomplete or incorrect pathway knowledge in simulations.
• Real metabolomics data analysis from the COPDGene study showed significant improvements in pathway association testing power.
• The method successfully identified meaningful improvements in detecting pathway associations.

Conclusions:

• PaIRKAT offers a powerful approach for identifying associations between biological pathways and phenotypes.
• The method effectively overcomes limitations posed by disconnected pathway structures in omics data.
• PaIRKAT's techniques are generalizable to other biological data with graph-like structures, aiding clinical research.