mRMRe

mRMRe implements an ensemble Minimum Redundancy Maximum Relevance (mRMR) feature-selection framework for identifying complementary and relevant features in high-throughput genomic data.

Key Features:

• Ensemble mRMR Technique: Extends traditional mRMR with an ensemble approach to explore the feature space and construct robust predictors with improved prediction accuracy.
• Parallelized Implementation: Accelerates computations using OpenMP parallel processing with performance-critical code implemented in C.
• Support for Multiple Variable Types: Performs feature selection for continuous, categorical, and survival variables.
• Mutual Information-based Selection: Computes mutual information matrices for different variable types to inform the mRMR selection process.

Scientific Applications:

• Genomic feature selection: Identification of biologically significant genes from high-throughput genomic datasets.
• Biomarker discovery: Selection of complementary features for constructing predictive biomarkers in studies of disease mechanisms.
• Predictor construction: Assembly of feature sets for building robust predictive models with improved accuracy over classical mRMR methods.

Methodology:

Computation of mutual information matrices for continuous, categorical, and survival variables followed by feature selection using an ensemble mRMR algorithm with OpenMP-parallelized C implementations.