Machine learning -driven discovery of Parkinsonian-like phenotypes in idiopathic REM sleep behavior disorder.

INTRODUCTION: This study harnessed AI-powered tools like principal component analysis (PCA) and multivariate models (PCA-MV) to integrate clinical and laboratory data from Parkinson’s disease (PD) patients, individuals with idiopathic REM sleep behavior disorder (iRBD), and healthy controls. The aim was to identify phenotypic similarities between iRBD and PD and to explore subgroup patterns within the iRBD population. METHODS: Baseline clinical and laboratory data were analyzed from 49 participants: PD (n = 15), iRBD (n = 19), and healthy controls (n = 14). Variables included Hoehn and Yahr scale, UPDRS Part III, MOCA, MMSE, SCOPA-cognitive, SCOPA-Autonomic Test, Beck Depression Inventory, Farnsworth Munsell 100 Hue Test, PSQI, ESS, RBDQ-HKQ, polysomnography/video-polysomnography and plasma microRNA 27/29 levels. PCA was used for dimensionality reduction, followed by supervised machine learning models, including logistic regression and support vector machines (SVM). Model performance was assessed using F1-scores. RESULTS: Supervised classification models effectively differentiated PD patients from healthy controls. The PCA–SVM approach achieved high performance (F1-score = 0.92), with radial basis function kernels providing the best separation. Using this model, 69% of iRBD patients clustered within the Parkinsonian feature space defined by shared clinical and biological characteristics. CONCLUSION: Multivariate machine learning combined with dimensionality reduction enables the integration of heterogeneous clinical and molecular data, improving the identification of similarities and subgroup patterns across neurological conditions. PCA-based models may be useful for stratifying iRBD patients according to their degree of similarity to PD.