Machine Learning Identifies TRAPPC13/COPS5 as Biomarkers and Vesicle Transport Subtypes in Parkinson's Disease.

BACKGROUND

Parkinson's disease (PD) is a neurodegenerative disorder characterized by neuron loss and abnormal protein trafficking. Dysregulation of vesicle-mediated transport contributes to pathogenesis, but its diagnostic value and immune associations are unclear.

METHODS

Transcriptomic data from GEO datasets (GSE20141, GSE20163, GSE7621) were analyzed. Differentially expressed vesicle-mediated transport-related genes were identified. Machine learning algorithms (least absolute shrinkage and selection operator, random forest, extreme gradient boosting) were integrated to select robust diagnostic biomarkers. The diagnostic model was validated across independent datasets. Immune infiltration was evaluated, and non-negative matrix factorization (NMF) identified molecular subtypes.

RESULTS

Machine learning revealed TRAPPC13 and COPS5 as robust diagnostic biomarkers with high predictive accuracy. The diagnostic model demonstrated strong accuracy across multiple datasets and showed excellent calibration and clinical applicability. Immune analysis highlighted differences in CD8+ T-cell fraction and MHC class I signaling between PD and controls. NMF clustering identified two transcriptionally distinct PD subtypes with distinct pathways and immune signatures.

CONCLUSION

This analysis identified TRAPPC13 and COPS5 as novel vesicle transport-related diagnostic biomarkers for PD. These genes show strong diagnostic potential, and the two identified molecular subtypes offer new insights into PD pathogenesis and may guide personalized therapeutic strategies.