Enhancement of gut-microbiome-based classification of prodromal Parkinson's Disease cases by hyperparameter tuning, integration of multimodal data, and meta-analysis.

Abstract

Parkinson's Disease (PD) is often preceded by a prodromal phase marked by non-motor symptoms such as REM Sleep Behaviour Disorder (RBD), gastrointestinal dysfunction, and olfactory impairment. Mounting evidence implicates gut microbiota in the early pathophysiology of PD. In this study, we aimed to enhance the classification of prodromal PD using gut microbiota profiles, with a focus on model performance and biomarker identification. Publicly available gut microbiome datasets of individuals with RBD and healthy controls were analyzed. Following batch correction and genus-level filtering, we identified six taxa significantly depleted in both RBD and PD groups, all belonging to short-chain fatty acid (SCFA)-producing genera. Hyperparameter tuning on the testing set achieved an AUC of 0.889. When combining microbial and clinical features, the model achieved a peak mean AUC of 0.917. Feature selection identified 11 top-ranked features, yielding an AUC of 0.849. Using only the six differentially abundant SCFA-producing taxa, the model retained a strong predictive capacity with a mean AUC of 0.853. These results demonstrate the utility of gut microbiota, particularly SCFA-producing taxa, in the early identification of prodromal PD. Integration with select clinical features further improves model performance, supporting their potential role in early diagnostic frameworks.