Beyond Methodological Noise: Robust Gut Microbiota Signals in Parkinson’s Disease via a Heterogeneous Multi-Algorithmic Framework

Abstract

Background: Recent research has highlighted the role of gut microbiota dysfunctions in Parkinson’s disease (PD), which may occur decades before the onset of motor symptoms. However, methodological variability across studies and complex host–microbiota interactions limit conclusions regarding which genera are most relevant to these dysfunctions. This study aimed to identify robust and reproducible microbial signatures associated with PD. Methods: To this end, we developed a consensus-based framework combining differential abundance analysis (DAA) with machine-learning analyses across gut microbiota datasets from individuals with PD and compared inferred functional profiles across cohorts. Results: Pairwise genus overlap ranged from 56 to 98 genera, with 67–75% concordance in the direction of association. The predictive model achieved AUCs of 0.75–0.95 within individual cohorts and a mean AUC of 0.72 under Leave-One-Cohort-Out validation (LOCO). Fourteen genera were consistently identified across all cohorts, including Porphyromonas in PD and Roseburia in controls, alongside patterns involving SCFA-associated microbial functions and inflammation-related taxa. Conclusions: These findings were consistent with previous reports and support the relevance of the gut–brain axis in PD, highlighting the potential of microbiota-derived features as candidate microbial signatures.