Early Parkinson’s Revealed by Unlocking Longitudinal Omics at Population Scale

Many diseases begin developing years before symptoms appear1-3, yet biospecimens from these early stages are rarely available. We developed Chronos, a framework that uses privacy-preserving tokenization4 to link archived plasma samples with longitudinal clinical records, enabling the modeling of molecular trajectories across time. Starting with >100 million archived, routine-donation samples from 3 million plasma donors, we assembled a longitudinal Parkinson’s disease cohort and profiled 2,609 samples from 348 cases and 348 matched controls using four proteomics platforms, covering more than 25,000 proteoforms. We reproduced proteomic signatures from clinically-phenotyped cohorts and revealed early, coordinated alterations in a CXCL12, cell adhesion, and integrin signaling network years before the estimated onset of PD. We used protein ratios to predict future diagnosis, achieving a maximum cross-validated area under the curve of 0.76 and replicated the findings in up to 5 independent cohorts. Chronos enables disease detection before clinical manifestation by prioritizing longitudinal molecular changes over symptoms, and provides a general framework to reconstruct chronic and acute disease trajectories from large plasma collections.