A BAYESIAN NONPARAMETRIC MIXTURE MODEL FOR SELECTING GENES AND GENE SUBNETWORKS.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput data analysis presents challenges in selecting informative features from numerous measurements.
• Existing gene selection methods often rely on parametric/regression models using gene network information.

Purpose of the Study:

• To propose a novel nonparametric Bayesian model for gene selection that incorporates network information.
• To identify genes strongly associated with clinical outcomes and those with unique expression patterns.
• To address limitations of regression models in capturing specific gene expressional behaviors.

Main Methods:

• Developed a nonparametric Bayesian model for gene selection integrating gene network information.
• Demonstrated the model's equivalence to an infinite mixture model.
• Implemented posterior computation using Markov chain Monte Carlo (MCMC) methods.
• Proposed two efficient algorithms for approximating posterior simulations with reduced computational cost.

Main Results:

• The proposed model effectively selects informative genes associated with clinical outcomes.
• The model successfully identifies genes exhibiting particular expressional behaviors, which are often missed by regression models.
• Simulation studies and analysis of yeast cell cycle microarray data validated the method's performance.

Conclusions:

• The nonparametric Bayesian model offers a powerful alternative for gene selection in high-throughput data analysis.
• The model's ability to capture diverse gene expressional patterns enhances its utility in biological research.
• Efficient computational algorithms facilitate the practical application of this method.