A Pan‐Cancer Microenvironment Atlas Constructed by DECEPTICONx

ABSTRACT The tumor microenvironment (TME) plays a pivotal role in shaping cancer progression, immune evasion, and therapeutic response. However, a systematic understanding of immune and stromal cell heterogeneity across cancer types—and their associations with clinical phenotypes—remains lacking due to limited single‐cell data and the associated clinical metadata. Here, we present a comprehensive pan‐cancer TME cell atlas to date, quantifying the composition of 102 immune and stromal subtypes across 30 cancer types and encompassing over 11,000 bulk RNA‐seq samples from TCGA. To enable this unprecedented resolution, we developed DECEPTICONx, a flexible computational framework that integrates single‐cell RNA‐seq data by constructing high‐resolution expression templates, which are then paired with context‐optimized deconvolution strategies, dynamically selecting the best‐matched template‐deconvolution combinations for each task. By leveraging 535,665 single cells spanning T cells, B cells, fibroblasts, macrophages, and NK cells, we constructed high‐resolution expression templates and systematically deconvolved bulk samples. The resulting atlas revealed robust and novel associations between specific cell subtypes and patient prognosis, immunotherapy response, and inter‐cellular coordination within the TME. In summary, the DECEPTICONx‐based atlas not only advances our understanding of tumor‐immune ecology but also offers a foundation for biomarker discovery and therapeutic development in precision oncology.