ExhauFS: exhaustive search-based feature selection for classification and survival regression.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Feature selection is crucial for preventing overfitting in machine learning.
• Exhaustive search is an effective but often unavailable feature selection method.
• Biomedical research lacks accessible implementations for exhaustive feature selection.

Purpose of the Study:

• To introduce ExhauFS, a user-friendly command-line tool for exhaustive feature selection.
• To demonstrate ExhauFS's utility in classification and survival regression tasks.
• To benchmark ExhauFS against state-of-the-art methods in real-world cancer datasets.

Main Methods:

• Exhaustive search algorithm implemented in ExhauFS.
• Application of ExhauFS for gene signature construction in breast cancer recurrence prediction.
• Utilizing ExhauFS for isomiR signature development in colorectal cancer survival analysis.
• Benchmarking ExhauFS against L1-regularized models.

Main Results:

• ExhauFS successfully constructed gene signatures for breast cancer with high sensitivity and specificity.
• Identified cross-platform gene signatures for breast cancer recurrence prediction.
• Developed predictive isomiR signatures for colorectal cancer survival with good concordance index.
• ExhauFS outperformed alternative methods in cross-platform validation and predictive accuracy.

Conclusions:

• ExhauFS provides a robust and accessible solution for exhaustive feature selection in biomedical research.
• The tool enables the discovery of reliable and cross-platform predictive signatures.
• ExhauFS demonstrates superior performance compared to existing feature selection approaches for cancer-related datasets.