Objective: Examine whether adolescent brain structural deviations contribute to Parkinson’s disease (PD) susceptibility.

Background: Genetic variants associated with PD have been linked to neurodevelopmental processes, and early-life environmental exposures have been proposed as potential risk factors[1]. Additionally, structural brain differences in individuals at risk for PD hint at a possible developmental component[2]. However, direct evidence remains limited. This study applies machine learning to investigate if polygenic risk scores (PRS) of adolescent brain structure deviations can provide insights into potential neurodevelopmental contributions to PD.

Method: We acquired adolescent brain structural MRI data for cortical thickness, surface area, and subcortical volume from the Adolescent Brain Cognitive Development study and calculated their deviation scores using normative modeling via CentileBrain, which estimates individual deviations from a normative reference population[3]. Brain regions were classified as supranormal and infranormal if their deviation scores fell outside the range of [-1.96,1.96][4]. For each brain region, we then conducted a genome-wide association study (GWAS) to identify genetic variants associated with deviations. Using the GWAS summary statistics, we computed PRS for 7 PD cohorts (13,988 PD cases and 27,203 controls, Fig. 1). We then implemented a two-step XGBoost modeling approach to predict PD status. In the first step, we trained an XGBoost model using all features to identify the most important contributors to PD prediction based on SHapley Additive exPlanations values. Finally, we refined the model with only the selected important features.

Results: The final XGBoost model achieved AUC = 87%, precision = 66%, recall = 89%. After adding PD PRS, the model performance achieved AUC = 90%, precision = 76%, recall = 86%. In addition, the model achieved similar but slightly lower performance with left fusiform surface area only before and after adding PD PRS. The genetic correlation between this region and PD was -0.54 (p > 0.05). The post-GWAS analysis of this region identified KCNMA1, a gene linked to dystonia and obesity [5], [6].

Conclusion: Our findings indicate that variants associated with adolescent brain structure deviations, may contribute to PD risk. This study provides novel insight to PD risk, underscoring the importance of brain development in PD.

Figure 1. workflow summary.

References: [1] K. L. Grasby et al., “The genetic architecture of the human cerebral cortex,” Science, vol. 367, no. 6484, p. eaay6690, Mar. 2020, doi: 10.1126/science.aay6690.
[2] N. Abbasi et al., “Neuroanatomical correlates of polygenic risk for Parkinson’s Disease,” Nov. 28, 2022, medRxiv. doi: 10.1101/2022.01.17.22269262.
[3] R. Ge et al., “Normative modelling of brain morphometry across the lifespan with CentileBrain: algorithm benchmarking and model optimisation,” The Lancet Digital Health, vol. 6, no. 3, pp. e211–e221, Mar. 2024, doi: 10.1016/S2589-7500(23)00250-9.
[4] ENIGMA Clinical High Risk for Psychosis Working Group, “Normative Modeling of Brain Morphometry in Clinical High Risk for Psychosis,” JAMA Psychiatry, vol. 81, no. 1, pp. 77–88, Jan. 2024, doi: 10.1001/jamapsychiatry.2023.3850.
[5] G. Zhang et al., “A Gain-of-Function Mutation in KCNMA1 Causes Dystonia Spells Controlled with Stimulant Therapy,” Mov Disord, vol. 35, no. 10, pp. 1868–1873, Oct. 2020, doi: 10.1002/mds.28138.
[6] H. Jiao et al., “Genome wide association study identifies KCNMA1 contributing to human obesity,” BMC Med Genomics, vol. 4, p. 51, Jun. 2011, doi: 10.1186/1755-8794-4-51.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/linking-adolescent-brain-development-to-parkinsons-disease-risk-using-machine-learning-on-polygenic-risk-scores/