Network-based genomic discovery: application and comparison of Markov random field models

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene networks are crucial for understanding biological systems.
• Existing mixture models for genomic data analysis have limitations.
• Markov random fields (MRFs) offer a promising approach for modeling gene-gene interactions.

Purpose of the Study:

• To compare the practical performance of Discrete- and Gaussian-Markov random field (DMRF/GMRF) mixture models for genomic data analysis.
• To introduce novel prior specifications for GMRF models and a Bayesian approach for DMRF models.
• To evaluate the accuracy of False Discovery Rate (FDR) estimation using posterior probabilities within MRF models.

Main Methods:

• Application of DMRF and GMRF mixture models to genomic data.
• Development of novel prior constraints for GMRF models.
• Implementation of a fully Bayesian approach for DMRF models.
• Assessment of FDR estimation accuracy via posterior probabilities.

Main Results:

• Modified GMRF models demonstrated superior performance compared to other methods.
• Both MRF-based mixture models outperformed the standard mixture model.
• MRF models exhibited robustness even with misspecified gene networks.
• The study assessed FDR estimation accuracy in the context of MRF models.

Conclusions:

• MRF-based mixture models represent a significant advancement in genomic data analysis.
• Modified GMRF models offer enhanced performance and robustness.
• These models provide a more accurate framework for understanding complex gene-gene interactions and managing FDR.