A data-driven approach to identify molecular subtypes and therapeutic targets in Parkinson’s disease

This study explores the potential of coupling unsupervised machine learning (ML) with bioinformatics and network analysis to elucidate the molecular landscape of Parkinson's disease (PD) subtypes, aiding in diagnosis and drug repurposing. We applied a hybrid datadriven workflow to gene expression data from idiopathic PD post-mortem brain samples, integrating multiple unsupervised ML algorithms for disease subtyping, clusterability assessment, and cluster determination. Network and bioinformatics analyses were used to identify common regulatory genes in specific disease networks. Key genes were tested in a drug repurposing pipeline, yielding compounds with disease-modifying potential. We then replicated the experiment on RNA-seq whole blood data, aiming to identify stable molecular subtypes based on gene expression. This approach offers a precision medicine strategy for PD, addressing the heterogeneity of the disease and advancing our understanding of its molecular underpinnings.