EXPLAINABLE AI-DRIVEN MULTICOHORT TRANSCRITOMIC PROFILING OF THE SUBSTANTIA NIGRA REVEALS BIOMARKERS FOR PARKINSON’S DISEASE

Abstract

The current landscape of Parkinson’s Disease (PD) biomarkers remains limited by inconsistent reproducibility, the limitation of traditional analytical frameworks, and cohort heterogeneity to connect statistical findings with underlying biological mechanisms. Explainable artificial intelligence (xAI) based methods open opportunities in biomarker discovery by combining predictive accuracy and biological explanation, thereby bridging the reproducibility and mechanistic limitations of conventional single cohort studies. In this study, seven transcriptome datasets from substantia nigra tissue retrieved from the NCBI GEO database were analysed to identify robust and biologically coherent PD-associated biomarkers. After the preprocessing step, a total of 433 commonly expressed genes were utilized in performing ML modelling techniques. Among four ML classifiers, XGBoost achieved the highest value for the area under the receiver-operating characteristic curve in determining the accuracy of its diagnostic tests, registering an AUC of 0.86. The SHAP value demonstrated that TARDBP, EIF4E, and SHQ1 were highly ranked genes and emerged as among the most influential biomarkers. Functional enrichment analysis continued to produce significant activation of RNA transport, oxidative phosphorylation, and prolactin signalling as prominent pathways in PD. The prolactin signalling established a reproducible and mechanistically interpretable molecular signature that modulates JAK/STAT, PI3K/AKT, and MAPK signalling pathways that are directly associated with microglial activation and oxidative and inflammatory stress responses. The study also suggests the capability of xAI to uncover endocrine-linked regulatory pathways that are often overlooked in traditional transcriptomic analyses, offering promising biological insights and potential therapeutic.