Ten protein panel predicts ALS from plasma with high accuracy

Predictive biomarkers are increasingly used to guide clinical trial design, diagnosis, and management of patients with neurodegenerative diseases. We profiled plasma from 632 ALS patients and 160 healthy controls using the NULISA CNS and Inflammation Panels and applied eXtreme Gradient Boosting (XGBoost) with nested cross-validation and SHapley Additive exPlanations (SHAP) to identify disease-associated proteins. A 12-protein candidate panel was refined using SHAP feature importance and absolute quantification, then evaluated in an independent validation cohort (144 ALS, 37 controls). We identified a final 10-protein panel (NEFL, FOLR1, pTau-181, TREM1, VEGFA, NEFH, CALCA, NRGN, LAMP3, and PTX3) that achieved high classification performance (balanced accuracy 98.3%; AUC 0.992). Model performance was driven by plasma proteomic signatures rather than clinical variables and was not dependent on neurofilament light chain alone. Representing over 750 ALS cases, this study supports development of a clinically deployable blood-based ALS screening tool. Using large-scale plasma proteomics, Kim et al. identify proteins altered in ALS and develop a 10-protein model with high classification accuracy, supporting development of a blood-based tool to facilitate earlier ALS diagnosis.