Identification of Parkinson’s Disease Associated Genes Through Explicable Deep Learning and Bioinformatic

Weutilized interpretable deep learning methodologies to discern critical genes and latent biomarkers associated with Parkinson’s disease (PD). Gene expression data were collected from the GEO dataset, subjected to rigorous differential expression analysis to curate genes for subsequent scrutiny. Based on the P-Net and PASNet models, we have developed a pathway-related deep learning model that integrates PD-associated gene expression data with established biological pathways. This method has yielded satisfactory results, manifesting an Area Under the Curve (AUC) of 0.73 and an F1 score of 0.71, thereby efficaciously discriminating PD patients and bestowing novel insights into the pertinent biological pathways. Through interpretable deep learning models, we have identified potential biomarkers (XK, PDK1, TUBA4B, TP53) and their associated biological pathways (innate immune system, hemostasis, G protein-coupled receptor signaling pathway) related to Parkinson’s disease. The importance of these genes has been validated through external datasets and UPDRS III scores. Of particular significance is the XK gene, also known as Kell blood group precursor, and numerous XK gene mutations have been linked to the McLeod syndrome which exhibits symptomatic similarities with PD. Taken together, we identified several PD associated genes by explicable deep learning and bioinformatics methods, and XK gene was demonstrated a close correlation to PD.