A Graph-Informed Modeling Framework Empowering Gene Pathway Discovery.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Gene expression data analysis is crucial for understanding biological processes and diseases.
• High-dimensional gene expression data presents statistical challenges in identifying significant pathways and classifying diseases.
• Integrating prior biological knowledge, such as gene regulatory networks, can improve analytical power.

Purpose of the Study:

• To develop a novel graph-informed modeling framework for statistical analysis of gene expression data.
• To enhance the identification of differentially expressed gene pathways.
• To improve gene expression-assisted disease classification in high-dimensional settings.

Main Methods:

• Integrated gene regulatory network information into hypothesis testing for mean vector differences and linear discriminant analysis.
• Employed a block-coordinate descent approach for linear structural equation modeling with mixed graph information (directed/undirected edges, cycles).
• Validated the framework through extensive simulations and application to a lung cancer dataset (Cancer Genome Atlas Program - TCGA).

Main Results:

• Demonstrated improved statistical power for gene pathway tests compared to existing methods.
• Showcased enhanced accuracy in gene expression-assisted disease classification.
• Successfully applied the graph-informed approach to identify differentially expressed gene pathways and classify lung cancer stages.

Conclusions:

• The graph-informed modeling framework effectively utilizes external gene interaction information for improved statistical analysis.
• Incorporating gene regulatory network information significantly enhances gene pathway analysis and disease classification.
• This approach holds potential for advancing gene pathway discovery, disease diagnosis, and treatment strategies.