Proteomics and machine learning identify a distinct biomarker panel to detect prodromal and early Parkinson’s disease

Abstract Parkinson’s disease (PD) is an increasingly prevalent neurodegenerative disease for which readily available and non-invasive diagnostic biomarkers are scarce. Here, we present a panel of proteomic plasma biomarkers, capable of discriminating between PD and healthy controls with 100% accuracy in a machine learning model. We performed a discovery proteomics study on newly diagnosed PD patients and controls, followed by a multiplexed targeted proteomic assay applied to 99 de novo PD patients and 36 controls. The machine learning model correctly classified all patients, and multiple markers correlated with motor, non-motor symptom severity and cognitive decline. We also evaluated 18 prodromal subjects with iRBD and predicted 72 - 94% of the iRBD samples as PD. This figure matches the clinical conversion rate observed in PD, identifying a pattern already evident in iRBD and indicating pre-symptomatic molecular events. These findings may advance our understanding and supporting of future clinical trials.