Identification of Alzheimer’s Disease Associated Genes through Explicable Deep Learning and Bioinformatic

We utilized interpretable deep learning to identify key genes and potential biomarkers associated with Alzheimer’s disease (AD). AD-related gene expression data from GEO datasets and AlzData were collected. To create a training set, we performed differential expression analysis to delete low-expressed genes. To explore disease-related genes and potential biomarkers, we employed a pathway-centric deep learning network, which integrated gene expression data with established biological pathways, building upon the foundation of PASNet. Our study employed a pathway-related deep learning model based on PASNet combined with bioinformatic analysis for AD risk prediction, and achieved a good performance (AUC = 0.82, F1 score = 0.73). The model accurately classified AD patients based on gene expression data from the brain tissue while providing interpretability deep neural networks. Potential biomarkers (DYNC1I1, DNAJC1, SCRIB, TFEB, MAPKAPK3) and biological pathways (Senescence-Associated Secretory Phenotype, Transcriptional activity of SMAD2/SMAD3:SMAD4 heterotrimer, Toll-like Receptor Cascades) associated with AD were identified. We have employed a combination of bioinformatics analysis and deep learning models to identify a series of candidate genes associated with the diagnosis of Alzheimer’s disease (AD). Moreover, we have validated their potential as biomarkers through the utilization of external datasets, and finally focus on the five genes DYNC1I1, DNAJC1, SCRIB, TFEB, and MAPKAPK3.