McTwo: a two-step feature selection algorithm based on maximal information coefficient.

Area of Science:

• Biomedical data analysis
• Bioinformatics
• Machine learning in healthcare

Background:

• High-throughput omics technologies generate vast, high-dimensional biomedical datasets.
• The
• paradigm, with many features and few samples, poses challenges for traditional analysis.
• Feature selection algorithms aim to identify relevant features associated with phenotypes to address this data imbalance.

Purpose of the Study:

• To introduce McTwo, a novel feature selection algorithm.
• To leverage the Maximal Information Coefficient (MIC) for feature selection.
• To evaluate McTwo's performance against existing algorithms in terms of classification accuracy and feature subset size.

Main Methods:

• Developed McTwo, a feature selection algorithm utilizing Maximal Information Coefficient (MIC).
• Designed McTwo to select features independently associated with phenotypes.
• Aimed to achieve high classification performance using the nearest neighbor algorithm with selected features.

Main Results:

• McTwo demonstrated comparable or superior performance to existing algorithms across 17 datasets.
• The algorithm significantly reduced the number of selected features while maintaining high classification accuracy.
• Features selected by McTwo showed notable biomedical relevance, supported by literature review.

Conclusions:

• McTwo provides a feature subset with excellent classification performance and a minimal feature count.
• The algorithm is a valuable complementary tool for analyzing high-dimensional biomedical data.
• McTwo offers a promising solution for the challenges posed by the
• data paradigm in biomedical research.