Objective: To integrate polygenic scores (PGS) for blood-based omics (gene expression, protein and metabolite levels) in machine learning models to improve the prediction of Parkinson’s disease (PD).

Background: Although genome-wide analyses of PD capture a substantial proportion of the twin heritability, PGS prediction remains limited. Molecular prediction models of PD based on other omics, such as proteomics, have previously shown promising results. However, genetic data is more widely available due to increasingly low costs. As omics are themselves subject to genetic influence, they can be proxied as PGSs to improve PD prediction beyond variables such as age, sex and total polygenic risk for PD.

Method: In European individuals aged >60 from the Global Parkinson’s Genetics Program (GP2) (Ncases = 21,493; Ncontrols = 7063) we calculated 10,512 blood based omics-PGS using SNP weights from OmicsPred. Associations omics-PGS and PD case/control status were tested using logistic regression, controlling for age, sex and the 10 principal components, with statistical significance defined as pFDR < 0.05.  Sensitivity analyses will control for total PD polygenic risk. Significant molecules will be tested for causality using Mendelian randomisation. Multi-omic, multi-PGS models will be trained using LASSO, elastic net and XGBoost in 70% of the European sample, and validated in the other 30%, with external testing in the UK Biobank, evaluating performance against baseline models of age, sex and PD polygenic risk. The generalisability of univariate associations and multi-omic, multi-PGS models will be assessed in all other available ancestries in GP2.

Results: In univariate analysis we identified 27 FDR significant associations between omics-PGS and PD case/control status, involving one metabolomic, four proteomic and 22 transcriptomic scores. The most significant positive association was observed for PRR14 expression (OR [95%CI] = 1.08 [1.05-1.11], pFDR = 9.15×10-5) and the most significant negative association for KANSL1-AS1 expression (OR [95%CI] = 0.89 [0.86-0.91], pFDR = 1.09×10-14). We expect multi-omic, multi-PGS models to improve PD prediction over age, sex and PD polygenic risk alone.

Conclusion: This study represents the first multi-omic, multi-PGS prediction modelling of PD, with the potential to improve risk prediction models, enhance our understanding of molecular pathways contributing to PD, and inform earlier intervention.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/multi-omic-multi-polygenic-score-prediction-of-parkinsons-disease/