Integrating Biological Knowledge Into Case-Control Analysis Through Iterated Conditional Modes/Medians Algorithm.

Area of Science:

• Biostatistics
• Computational Biology
• Statistical Genetics

Background:

• High-throughput technology necessitates efficient methods for high-dimensional logistic regression.
• Traditional logistic regression faces challenges with large datasets and complex variable interactions.
• Variable selection is crucial for building interpretable and predictive models.

Purpose of the Study:

• To develop a fast and efficient empirical Bayes logistic regression model for high-dimensional data.
• To enhance variable selection power by incorporating biological knowledge using Ising priors.
• To introduce the Iterated Conditional Modes/Medians (ICM/M) algorithm as a computationally advantageous alternative to MCMC.

Main Methods:

• An empirical Bayes logistic regression model utilizing spike-and-slab priors for variable selection.
• Incorporation of biological knowledge via Ising priors to improve selection power.
• Development and application of the Iterated Conditional Modes/Medians (ICM/M) algorithm for model fitting.

Main Results:

• The ICM/M algorithm demonstrates computational advantages over Markov Chain Monte Carlo (MCMC) methods.
• Simulation studies indicate that ICM/M outperforms benchmark methods (lasso, adaptive lasso) in reducing false positives.
• The proposed method shows competitive predictive ability and flexibility in variable selection based on local posterior probabilities.

Conclusions:

• The empirical Bayes logistic regression model with ICM/M algorithm provides an effective and efficient approach for high-dimensional data analysis.
• Incorporating biological knowledge via Ising priors enhances variable selection in complex datasets.
• The method offers a flexible framework for identifying important variables while controlling the false discovery rate, as demonstrated in a Parkinson's disease study.