Decoding mitochondrial dysfunction in Parkinson's disease: machine learning identification of genomic biomarkers in Drosophila melanogaster

Mitochondrial dysfunction is a hallmark of Parkinson’s disease (PD), and Drosophila melanogaster serves as a tractable model to investigate underlying molecular mechanisms. This exploratory study aimed to identify mitochondrial gene expression signatures associated with Parkinsonism in mutant and wild-type flies and evaluate their potential as predictive biomarkers using machine learning approaches. Publicly available RNA-seq and microarray datasets of PD-associated mutants (PINK1, parkin, DJ-1) were harmonized and preprocessed to generate a unified gene expression matrix. Differential expression analysis was performed to identify transcriptional perturbations, and supervised classifiers (Random Forest, Support Vector Machine, XGBoost) were applied to distinguish mutants from wild-type samples. Model performance was assessed using leave-one-out cross-validation (LOOCV). Differential expression analysis revealed downregulation of CG15308, CG34349, and msl-3, and upregulation of Blmp-1, GABA-B-RI, and CG1360 in mutant flies. Random Forest identified PH4alphaNE2, IntS8, and rod as top candidate biomarkers. Predictive accuracy was modest (best: 50%) due to limited dataset size. Despite limited predictive performance, this study demonstrates the feasibility of integrating mitochondrial transcriptomic data with machine learning to identify candidate genomic biomarkers in Drosophila PD models. The identified genes provide a foundation for experimental validation and future multi-omics studies, highlighting a novel approach for biomarker discovery in neurodegenerative disease models.