A Machine Learning Pipeline for Multiple Sclerosis Biomarker Discovery: Comparing explainable AI and Traditional Statistical Approaches

Explainable AI and traditional statistics find different—but complementary—biomarkers: The study found that SHAP (an explainable AI method) and classical statistical analysis identified both overlapping and unique genes associated with MS, suggesting that combining these approaches reveals insights that either method alone might miss.

AI-identified biomarkers are biologically meaningful: The genes identified through SHAP were validated through enrichment analysis and linked to known MS-related biological pathways (sphingolipid signaling, immune cell differentiation, and Epstein-Barr virus infection), demonstrating that explainable AI methods can discover scientifically relevant findings, not just statistical patterns.

Explainable AI makes machine learning more trustworthy for medical research: By using SHAP to understand why the machine learning model made its predictions, researchers could identify specific genes driving the model's decisions, making the AI system transparent and interpretable rather than a "black box"—a critical requirement for responsible use in healthcare applications.

Combining multiple datasets and methods strengthens biomarker discovery: The pipeline integrated eight publicly available datasets and compared two analytical approaches, suggesting that robust biomarker discovery requires both large-scale data integration and methodological triangulation rather than relying on a single analysis method.