Category: Parkinson's Disease: Genetics
Objective: To assess whether artificial intelligence could predict GBA1-mutated genotype in Parkinson’s Disease (GBA1-PD) patients according to the different impact of significant clinical features.
Background: GBA1 is the major genetic risk factor for PD [1]. Artificial intelligence (AI) is becoming increasingly helpful to assist humans in analyzing complex data sets. Particularly, machine learning has been applied to a wide range of disciplines, including genetics.
Method: A consecutive cohort of GBA1-PD patients was matched with a cohort of non-mutated PD patients (NM-PD) according to age, sex, disease duration, Hoehn & Yahr stage and Charlson Comorbidity Index. Patients underwent clinical assessment, including the MDS-UPDRS total scores and subscores, and the Montreal Cognitive Assessment. Patients are classified on a binary target (GBA1-PD or NM-PD). Chi-square and Mann-Whitney test was applied to identify features that significantly differ between the two groups. XGBoost was identified as the most suitable machine learning model for this supervised classification task. For testing, we employed a Leave One Out method. Additionally, SHAP was used for examining the contribution of each individual features in the predictive model in order to identify the most significant factors influencing the genetic categorization and the model confidence.
Results: The dataset comprised 116 patients: 58 GBA1-PD (males: 31; age: 64.47 years; disease duration: 7.95 years, MDS-UPDRS III: 28.95; CCI: 2.48) and 58 NM-PD (age: 64.64 years; disease duration: 7.67 years, MDS-UPDRS III: 35.98; CCI: 2.67). For each patient 124 distinct features were recorded. Among these features, 16 were detected as significantly different between these groups, 5 of which had been used by the model to identify mutated or non-mutated GBA1 genotype with an accuracy over 70%. These variables included family history, score and subscores for motor impairment (MDS-UPDRS 3.14, MDS-UPDRS 3.8 a-b and rigidity subscores) and cognitive impairment.
Conclusion: Our results confirm that AI could represent a promising approach in predicting the GBA1 mutated status in PD, underlying the potential role of AI in enhancing genetic screening and personalized medicine in PD.
Abstract previously presented at the 10° Società Italiana Parkinson e Disordini del Movimento/LIMPE-DISMOV ETS Congress; Milan, Italy; 11 April 2024.
References: [1] V. Yahya, A. Di Fonzo, and E. Monfrini, ‘Genetic Evidence for Endolysosomal Dysfunction in Parkinson’s Disease: A Critical Overview’, Int. J. Mol. Sci., vol. 24, no. 7, p. 6338, Mar. 2023, doi: 10.3390/ijms24076338.
To cite this abstract in AMA style:
G. Di Rauso, A. Ghibellini, S. Grisanti, F. Cavallieri, V. Fioravanti, E. Monfrini, G. Toschi, G. Portaro, J. Rossi, R. Sabadini, L. Gherardini, C. Lucchi, G. Biagini, L. Bononi, M. Gabbrielli, A. Di Fonzo, F. Valzania. Artificial intelligence: a potential predictor of GBA1-mutated genotype in Parkinson’s Disease patients? [abstract]. Mov Disord. 2024; 39 (suppl 1). https://www.mdsabstracts.org/abstract/artificial-intelligence-a-potential-predictor-of-gba1-mutated-genotype-in-parkinsons-disease-patients/. Accessed October 12, 2024.« Back to 2024 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/artificial-intelligence-a-potential-predictor-of-gba1-mutated-genotype-in-parkinsons-disease-patients/