MS-MLB benchmark hits 0.989 AUC for blood-based MS classification

MS-MLB (Multiple Sclerosis Machine Learning Benchmark) is a new open benchmark from researchers Adam Simson, Ankush Dutta, and Quang Bui, designed to standardize machine learning research for classifying multiple sclerosis from whole blood RNA expression data. Built on the public GSE17048 cohort, it frames an MS versus healthy control task and runs a shared, leakage-controlled evaluation pipeline that researchers can rerun without reconfiguring. The pipeline covers nested cross-validation, an untouched stratified holdout set, bootstrap confidence intervals, ROC and precision-recall analysis, calibration measurement, and a novel MS Research Score.

In the final benchmark summary, Gradient Boosting ranked first on the holdout set with an MS Research Score of 93.83, an AUC-ROC of 0.989, sensitivity of 0.950, specificity of 0.778, F1 of 0.927, and a Brier score of 0.050. The authors emphasize that this is not a clinically validated diagnostic tool—a blood RNA classifier cannot replace clinical diagnosis—but rather a reproducible baseline for researchers. To their knowledge, MS-MLB is the first open benchmark for this specific classification task with a documented external model submission pathway. The framework, code, and data are openly accessible, allowing other teams to submit models and compare results under identical conditions.