An integrated approach for the analysis of biological pathways using mixed models

Area of Science:

• Bioinformatics
• Statistical Genetics
• Systems Biology

Background:

• Gene set enrichment analysis (GSEA) is crucial for interpreting microarray data by testing coordinated gene changes.
• Existing methods like Gene Ontology (GO) and KEGG Pathway analysis integrate gene annotation databases for system-level insights.
• Combining weak signals from individual genes within pathways enhances statistical power.

Purpose of the Study:

• To propose and evaluate an alternative gene class testing approach using mixed models.
• To address limitations of existing methods in handling complex experimental designs and borrowing strength across genes.
• To compare the proposed mixed models approach with established methods like GSEA and PAGE.

Main Methods:

• Development of a gene class testing framework based on mixed statistical models.
• Modeling and leveraging information across genes within pathways, including both up- and down-regulated genes.
• Comparison with non-parametric (GSEA) and parametric (PAGE) methods via simulation studies.
• Application to real-world datasets, including diabetes and dose-response studies.

Main Results:

• The mixed models approach demonstrates improved statistical power under normal location-based alternative hypotheses.
• This method effectively handles complex experimental designs where permutation resampling is challenging.
• The framework allows for direct control of false positive and false discovery rates.
• Performance was validated through simulations and application to biological datasets.

Conclusions:

• Mixed models provide a robust and powerful alternative for gene set enrichment analysis in microarray studies.
• This approach enhances the ability to detect subtle, coordinated biological signals.
• The flexibility and statistical rigor make it suitable for complex genomic data analysis.