Objective: To integrate genome-wide association study summary statistics, expression quantitative trait loci (eQTL), individual-level RNA sequencing data and machine-learning approaches to identify molecularly informed cross-trait associations with Parkinson’s disease (PD) and improve risk prediction.

Background: Thousands of genes are differentially expressed in the blood of individuals with PD. Blood-based polytranscriptomic scores (PTS) – the sum of observed gene expression weighted by transcriptome-wide association study (TWAS) effects – can provide a parallel approach to polygenic scores for investigating PD biology and prediction.

Method: We calculated 553 PTS for 100 phenotypes, including PD, using TWAS/summary-based Mendelian randomisation (SMR) associations and baseline RNA-seq data from three AMP-PD cohorts (Parkinson’s Disease Biomarkers Program (PDBP), Parkinson’s Progression Markers Initiative (PPMI), and Harvard Biomarkers Study (HBS); Ncases = 1,644; Ncontrols = 1,087). In each, PTS associations with PD case-control status were tested using logistic regression adjusted for age, sex, and surrogate variables, followed by fixed-effects meta-analysis. Meta-analysed PTS passing a Bonferroni threshold (p < 9 × 10⁻⁵) underwent sensitivity analyses using conditionally independent TWAS/SMR genes, followed by gene set analysis, Mendelian randomisation (MR), and cluster-based MR. Multi-PTS models combining scores using LASSO, elastic net, and XGBoost were trained in PDBP and externally tested in PPMI and HBS, with performance compared to an age+sex model.

Results: In total, 32 PTS were associated with PD, 26 remaining significant after sensitivity analyses. Positive associations were observed for neurological disorders including Alzheimer’s disease, essential tremor, and Lewy body dementia (LBD). MR supported a risk-increasing causal effect of PD on LBD driven by an 11-instrument cluster. Multi-TRS models achieved AUCs of 0.62–0.68 in HBS, outperforming age+sex (AUC [95%CI] = 0.54 [0.49–0.58]). Performance was limited in PPMI (age+sex AUC [95% CI] = 0.59 [0.55–0.62]; multi-TRS AUC range = 0.59–0.63).

Conclusion: This study represents the first large-scale cross-trait PTS analysis of PD and the first application of multi-PTS models for prediction. Findings indicate that PTS can identify novel cross-trait associations with PD and that multi-PTS models have the potential to improve prediction beyond demographic variables.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/cross-trait-and-multi-polytranscriptomic-score-prediction-of-parkinsons-disease/