Abstract
<title>Abstract</title> <p>As the number of clinical, imaging, and proteomic biomarker tests increase for Parkinson's disease (PD), determining which biomarkers and their coordination patterns drive disease becomes more difficult. This paper introduces PRISM-PD (Perturbation-based Resolution of Inter-biomarker Synchrony Mapping for Parkinson's Disease), a computational framework based on Hamiltonian spectral decomposition of multimodal biomarker covariance structure. PRISM-PD was applied to longitudinal data from 4,649 participants in the Parkinson's Progression Markers Initiative (PPMI), integrating dopamine transporter imaging, clinical assessments, and cerebrospinal fluid (CSF) neuroinflammatory proteomics. Unlike black-box deep learning architectures, this lightweight method runs on standard CPU hardware in seconds with full per-feature saliency, using a pre-specified 70/30 hold-out validation throughout. Eigendecomposition revealed fragmentation of neuroinflammatory coordination at PD diagnosis, driven by CSF proteins (CCL2, IL-6ST, CXCL13) with near-zero dopaminergic loading, supporting a two-axis model of early PD biology in which inflammatory coupling disruption and nigrostriatal degeneration are independent processes. Longitudinal analysis showed progressive coupling concentration whose eigenvector transition sequence parallels the clinical progression described by Braak and colleagues. Tremor-dominant and PIGD-dominant phenotypes showed distinct coupling trajectories, and CSF IL-6ST differed nominally across LRRK2-PD, GBA-PD, and sporadic PD genotypes. PRISM-PD is a lightweight, explainable, and deployable computational tool for quantifying multimodal biomarker coordination in PD, with direct application to genotype-stratified trial design and a generalizable framework for other neurodegenerative diseases.</p>