Leveraging machine learning to predict Parkinson's disease using pre-symptomatic proteomics data

Abstract In Parkinson’s disease, non-motor symptoms precede the characteristic motor manifestations by up to 20 years. However, predicting the disease risk remains challenging. We analyzed 2,937 plasma proteins assayed in UK Biobank participants with Parkinson’s disease, and applied machine learning (ML) to predict disease outcome up to 14 years before diagnosis. 446 proteins were dysregulated in incident cases compared to controls. A subset of 23 proteins predicted disease with an AUC of 0.78 in incident and 0.795 in prevalent cases. We validated the utility of the identified proteomic signatures in an independent cohort on 16 proteins shared across datasets, leading to a validation AUC up to 0.76. We identified overrepresentation of Parkinson’s disease-related pathways, including enrichment of Neuron Death and Amyloid-beta Clearance, up to 9 years before the diagnosis. Co-expression network analysis revealed protein modules associated with the disease risk and time-to-diagnosis. Parkinson’s disease signatures are found in plasma up to 14 years prior to diagnosis, with dysregulated proteins offering insights into disease development. Our findings underscore the potential of applying ML to large-scale proteomics data from pre-symptomatic patients for early disease prediction.