RCE-IFE: recursive cluster elimination with intra-cluster feature elimination.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Biology

Background:

• High-dimensional biological data presents computational and interpretational challenges.
• Feature selection (FS) is crucial for dimensionality reduction.
• Feature grouping is a foundational technique for effective FS.

Purpose of the Study:

• To propose a novel feature selection method, Recursive Cluster Elimination with Intra-Cluster Feature Elimination (RCE-IFE).
• To assess RCE-IFE's dimensionality reduction and discriminatory capabilities on diverse biological datasets.
• To evaluate the biological relevance and consistency of features selected by RCE-IFE.

Main Methods:

• Developed RCE-IFE, a supervised method that iterates feature grouping and elimination steps.
• Evaluated RCE-IFE on gene expression, miRNA expression, methylation, and metagenomics datasets.
• Compared RCE-IFE against various state-of-the-art FS methods and domain-specific tools.

Main Results:

• RCE-IFE achieved an average Area Under the Curve (AUC) of 0.85 on expression datasets with minimal features and shortest runtime.
• Outperformed several established FS methods (MRMR, FCBF, IG, CMIM, SKB, XGBoost) with an average AUC of 0.76 on gene expression data.
• Demonstrated comparable accuracy to Multi-stage on cancer datasets while using fewer features and showed high consistency in selected features.

Conclusions:

• RCE-IFE provides robust classifier performance and significantly reduces feature size.
• The method effectively maintains feature relevance and consistency across multiple runs.
• RCE-IFE offers a powerful solution for analyzing high-dimensional biological data.