A machine learning heuristic to identify biologically relevant and minimal biomarker panels from omics data

Area of Science:

• Biomedical data science
• Machine learning in omics

Background:

• Omics techniques generate vast, high-dimensional datasets.
• Feature selection is crucial for reducing data complexity while maintaining predictive power.
• Identifying small, accurate biomarker panels is key for clinical utility.

Purpose of the Study:

• To propose a novel heuristic, Rule-guided Iterative Feature Elimination (RGIFE), for selecting very small feature subsets from omics data.
• To identify potential biomarkers for osteoarthritis, articular cartilage degradation, and synovial inflammation.

Main Methods:

• RGIFE employs an iterative feature elimination process guided by rule-based machine learning.
• The heuristic was applied to both proteomic and transcriptomic datasets.
• Performance was compared against other feature selection methods and no feature selection.

Main Results:

• RGIFE enhanced classification accuracies across all tested datasets.
• The method successfully reduced datasets to a smaller number of relevant genes or proteins.
• Identified features included known markers for osteoarthritis and joint inflammation.

Conclusions:

• RGIFE is a suitable feature reduction method for both proteomic and transcriptomic data analysis.
• Feature reduction in omics data is beneficial for rheumatology, potentially improving diagnosis, treatment, and drug discovery.