An Intrinsic-hoc Framework for Heterogeneous Cellular Senescence Elucidation Using Deep Graph Representation Learning and Experimental Validation

Characterizing senescent cells and identifying corresponding senescence-associated genes within complex tissues is critical for our understanding of aging and age-related diseases. We present DeepSAS, an intrinsic-hoc framework for elucidating the heterogeneity encoded in senescent cells and their associated genes from single-cell RNA-seq data using deep graph representation learning. Applied to both healthy eye cell atlas and in-house idiopathic pulmonary fibrosis datasets with Xenium spatial transcriptomics validation, DeepSAS reveals robust and biologically grounded senotypes and demonstrates superior benchmarking performance compared with existing methods.