Objective: The aim of this study is to explore whether artificial intelligence (AI) can accurately cluster Parkinson’s Disease (PD) patients carrying different variants in PD-associated genes by considering only selected clinical features and not their gene mutations.

Background: Monogenic forms of PD account for approximately 5-10% of PD cases. Pathogenetic variants in PRKN and LRRK2 genes are the most common cause of early-onset PD and of monogenic PD, respectively. GBA1 is the major genetic risk factor for PD. PRKN-PD, LRRK2-PD and GBA1-PD patients often present characteristic clinical features [1].

Method: PD patients carrying pathogenetic variant in GBA1, LRRK2 and PRKN genes were included in this study. We collected demographic and clinical variables at PD onset and at 5 years follow up, including motor and non-motor PD symptoms, H&Y score, Levodopa Equivalent Daily Dose (LEDD) and presence of medical comorbidities. Chi Square and Mann-Whitney Tests were applied to select features that significantly differ between the three groups, identifying 12 key variables. Principal Component Analysis (PCA) was applied to reduce dimensionality from 12 to 10 variables, preserving 98% of the explained variance. K-Means clustering (k=3) was then performed on the PCA-transformed data, revealing three distinct clusters. Finally, t-distributed stochastic neighbor embedding (t-SNE) was used for visualization.

Results: 112 PD patients were included: 70 GBA1-PD, 23 LRRK2-PD and 19 PRKN-PD. The key variables identified were age at PD onset, presence of dystonia and bradykinesia at onset, motor phenotype at 5 years follow-up (presence of bradykinesia, rigidity or dystonia), dysautonomia, hallucination or cognitive impairment at 5 years from onset, LEDD and H&Y scale at 5 years follow up. The clustering showed that 71% of GBA1-PD belonged to cluster 1, 63% of PRKN-PD were in cluster 2, and 52% of LRRK2-PD were in cluster 3. The visualization demonstrated a separation between genetic groups.

Conclusion: Our results suggest that AI can identify meaningful clusters, reflecting the different PD genetic subtypes, by considering selected clinical features.

References: [1] Jia, F.; Fellner, A.; Kumar, K.R. Monogenic Parkinson’s Disease: Genotype, Phenotype, Pathophysiology, and Genetic Testing. Genes 2022, 13, 471

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/artificial-intelligence-unveils-genetic-etiologies-in-parkinsons-disease-cases-clustered-by-clinical-features/