Fig. 1: Neurotransmitter receptor-enriched multifactorial causal modeling.

a Each patient’s longitudinal pathological progression is decomposed into local effects due to: (i) direct influence of every imaging-derived biological factor (e.g., atrophy on resting state functional activity), (ii) receptor density distribution (e.g., D₁ receptor density on DAT loss), and (iii) receptor-pathology interactions (e.g., D₁ receptors × DAT interactions on functional activity), in addition to (iv) network-mediated inter-region propagation. Combining this data across (N_ROI = 95) brain regions and multiple visits results in a multivariate regression problem to identify the patient-specific parameters {α}. b Decomposing the covariance matrix of patients’ model-derived biological mechanism weights and clinical scores (specifically, the rates of decline of composite clinical scores; Methods: Clinical scores) identifies multivariate axes of receptor-factor interactions that are robustly correlated with the severity of combinations of clinical symptoms in PD (Methods: Biological parameters and relationship with cognition). c The regional contributions of receptor interactions to neurobiological changes are estimated by a feature importance analysis. We fit individualized models for every biological factor with and without each receptor map and performed permutation tests on the improvement in regional model residuals due to the inclusion of receptor maps. The resulting improvements are the significant regional influence of receptors on each target biological factor model.