Machine Learning for analysis of Multiple Sclerosis cross-tissue bulk and single-cell transcriptomics data

Machine learning models achieved high accuracy in identifying MS patients, with the cerebrospinal fluid B-cell model reaching an AUC of 0.94, demonstrating that AI can effectively distinguish disease signatures in biological data relevant to clinical diagnosis.

Explainable AI (SHAP) revealed disease mechanisms that traditional statistical methods missed, identifying immune checkpoint genes like ITK, CLEC2D, and KLRG1 that weren't highlighted by conventional differential expression analysis, showing how interpretability tools can uncover novel biological insights.

The study integrated multiple data types and sources (bulk microarray, single-cell RNA sequencing, blood, and cerebrospinal fluid samples), demonstrating responsible AI practice by validating findings across diverse datasets rather than relying on a single data source, which strengthens confidence in results.

Specific biological pathways were identified as potential MS drivers, including immune activation, Epstein-Barr virus-related pathways, and protein degradation processes, providing concrete mechanistic hypotheses that could guide future MS research and treatment development.

The research prioritized interpretability alongside performance, using SHAP analysis to explain why the model made predictions rather than treating it as a "black box," which is essential for responsible AI in healthcare where clinicians and patients need to understand the reasoning behind AI-assisted decisions.