Explainable Gated Recurrent Unit with Hybrid Attention and Memory-Augmented Network for Cell Types Classification in Alzheimer’s Disease Using Single-Nucleus Transcriptomics

Alzheimer’s Disease (AD) is characterized by complex cellular changes in the brain, also known as the most common form of dementia in late ages. Understanding these changes at the cellular level, particularly in the middle temporal gyrus (MTG), is crucial for developing targeted therapeutic strategies. In this study, we developed an Explainable Gated Recurrent Unit (GRU) model enhanced with hybrid attention and memory-augmented network (xGRAM) model to classify the cell types in MTG associated with AD. We employed single-nucleus RNA sequencing (snRNA-seq) data from the Seattle Alzheimer’s Disease Brain Cell Atlas (SEA-AD) to profile the gene expression of 49,043 single-nucleus transcriptomes from human MTG samples. The experiment revealed distinct gene expression patterns across the 24 (subclasses) cell types in MTG, highlighting their unique roles in AD pathology. The proposed xGRAM model achieved a high prediction accuracy of 98.76%, effectively identifying major cell types like L2/3 IT, L5 IT, L4 IT, Sst, Vip, Pvalb, and others. This study provides a comprehensive understanding of different cell types in MTG which are linked to neuroinflammation and synaptic dysfunction in AD. The findings suggest potential targets for therapeutic intervention, emphasizing the importance of cellular heterogeneity in AD research.