iMOMLogit

iMOMLogit implements a Bayesian variable-selection method for predicting binary responses in genomic datasets by using mixtures of non-local prior densities and point masses on regression coefficients to identify predictive explanatory variables in ultrahigh-dimensional settings.

Key Features:

• Bayesian Framework: Uses a Bayesian approach incorporating a mixture of non-local prior densities and point masses on binary regression coefficient vectors.
• Improved Model Identification: Demonstrates superior ability to identify the correct underlying model and reduce estimation and prediction errors, as shown in simulation studies.
• Efficient Variable Selection: Selects a minimal set of highly predictive explanatory variables to achieve high prediction accuracy with fewer predictors.
• Prior Hyperparameter Calibration: Sets prior hyperparameters by examining the total variation distance between the priors on regression parameters and the distribution of the maximum likelihood estimator under the null hypothesis.
• Ultrahigh-Dimensional Algorithm and Diagnostics: Provides a computational algorithm tailored for ultrahigh-dimensional settings and diagnostics to evaluate the probability of having identified the highest posterior probability model.

Scientific Applications:

• Genetic association studies: Variable selection and model identification for detecting associations between genomic predictors and binary phenotypes.
• Disease outcome prediction: Construction of parsimonious predictive models for binary clinical outcomes using genomic data.
• Model simplification and interpretation: Reduction of model complexity to facilitate clearer interpretation and validation of genomic findings.

Methodology:

Implements a Bayesian model-selection framework with a mixture of non-local priors and point-mass components on regression coefficients; prior hyperparameters calibrated via total variation distance to the null MLE distribution; employs a computational algorithm for ultrahigh-dimensional settings with diagnostics and evaluates performance through simulation studies.