Artificial Intelligence Predicts GBA1 Mutated Status in Parkinson's Disease Patients

Area of Science:

• Neurogenetics
• Computational Biology
• Medical Informatics

Background:

• Genetic variants in GBA1 are a primary risk factor for Parkinson's Disease (PD), contributing to 5-30% of cases.
• Identifying GBA1-mutated PD (GBA1-PD) is crucial for understanding disease mechanisms and potential targeted therapies.

Purpose of the Study:

• To evaluate the efficacy of Artificial Intelligence (AI) in predicting GBA1-mutated genotype in Parkinson's Disease patients.
• To develop a Machine Learning (ML) model for pre-test estimation of GBA1-mutated status using clinical and demographic data.

Main Methods:

• A cohort study compared 58 GBA1-PD patients with 58 non-mutated PD (NM-PD) patients.
• 124 features were analyzed using a supervised classification task with XGBoost ML model.
• Leave-One-Out cross-validation and SHapley Additive exPlanations (SHAP) were employed for model testing and feature analysis.

Main Results:

• An AI model accurately predicted GBA1-mutated genotype with 73% accuracy, based on four clinical features.
• Prediction accuracy reached 94% in a subset of patients with high SHAP confidence levels.
• Key predictive variables included family history and specific scores for cognitive and motor impairment (MDS-UPDRS).

Conclusions:

• AI demonstrates significant potential for improving targeted genetic screening in Parkinson's Disease.
• The developed model can assist in clinical settings, particularly where resources for genetic testing are limited.
• Further validation on larger, independent cohorts is necessary to refine the predictive model.