PMMLogit

PMMLogit performs Bayesian hierarchical logistic modeling for binary phenotype classification and genomic predictor selection in high-dimensional genomic datasets.

Key Features:

• Bayesian Hierarchical Modeling: Incorporates prior information and quantifies uncertainty in model parameters for high-dimensional genomic data.
• Polya-Gamma Data Augmentation: Uses Polya-Gamma augmentation to address non-conjugacy between logistic likelihoods and common priors and to enable efficient posterior computation.
• Markov Chain Monte-Carlo (MCMC) Techniques: Employs MCMC sampling to obtain exact posterior samples for parameter inference in high-dimensional settings.
• Model Selection via Posterior Probabilities: Evaluates posterior model probabilities to identify combinations of genomic predictors associated with binary phenotypes.
• Bayesian Model Averaging (BMA): Implements BMA using the posterior mean to average predictions across models visited during MCMC sampling.

Scientific Applications:

• Genomics research: Applied to high-dimensional genomic datasets for inference and prediction with binary phenotypic outcomes.
• Genomic predictor selection: Selects true genetic models and relevant genomic predictors in settings with many candidate variables.
• Phenotype classification and prediction: Improves estimation and predictive accuracy for binary phenotype classification relative to comparator methods.
• Disease-associated gene identification: Identifies significant genes associated with disease outcomes in complex diseases, including colon cancer and leukemia.

Methodology:

Uses a Bayesian hierarchical model for logistic regression with Polya-Gamma data augmentation, MCMC sampling for posterior inference, model selection based on posterior model probabilities, and Bayesian model averaging via the posterior mean.