DECEPTICONx maps immune and stromal cell diversity across 30 cancer types

Researchers have developed DECEPTICONx, a computational framework that maps the cellular diversity surrounding tumors by combining single-cell reference data with large collections of bulk RNA sequencing samples. Published in iMetaMed on January 17, 2026, the study presents a pan-cancer microenvironment atlas designed to help researchers investigate how immune and stromal cell populations relate to patient survival and immunotherapy response.

The tumor microenvironment contains interacting immune and support cells that can influence cancer behavior. Single-cell sequencing can distinguish these populations in detail, but its cost, technical demands and limited clinical follow-up can restrict large-scale analyses. Bulk RNA sequencing datasets often cover more patients and include survival or treatment information, although their measurements combine signals from many different cell types. Computational deconvolution uses reference expression profiles to estimate the contributions of these cell populations.

DECEPTICONx adapts the pairing of reference-template construction and deconvolution methods to each dataset. The framework evaluates three template-building algorithms and eleven deconvolution approaches. Using 535,665 single cells spanning T cells, B cells, fibroblasts, macrophages and natural killer cells, the team assembled reference profiles for 102 annotated subtypes. They then analyzed more than 11,000 samples from The Cancer Genome Atlas across 30 cancer types, connecting estimated cell compositions with available clinical information.

The atlas revealed subtype-specific associations that would be obscured by broad cell categories. For example, higher abundance of the B.c11.pre.GC B-cell subtype was associated with poorer survival across multiple cancers, contrasting with an earlier favorable interpretation. The study also found different survival patterns for fibroblast populations. Fibroblast.c04.LRRC15 was more abundant in immunotherapy non-responders in an analyzed cohort, supporting further investigation of its relationship with treatment resistance.

In melanoma immunotherapy datasets, the researchers used cell-composition patterns to define distinct immune ecotypes. The CD4.c23.Mix.NME1 subtype was enriched among non-responders and correlated with other populations associated with resistance. These analyses connect individual cell subtypes with broader cellular networks, providing hypotheses about coordinated immune suppression rather than establishing that any one subtype causes resistance.

The authors identify important limitations. DECEPTICONx relies on transcriptomic data and does not capture the spatial arrangement of cells within a tumor. Associations between estimated cell populations, prognosis and treatment response also require experimental work to clarify their underlying mechanisms. Integrating spatial transcriptomics and further biological validation could help explain which cellular relationships are functionally important.

The resulting atlas offers a resource for exploring tumor ecosystems across cancer types. Its contribution is to connect detailed cellular reference information with the scale and clinical context of existing bulk datasets, supporting future biomarker research and investigation of immune and stromal interactions.