Evaluating deconvolution methods using real bulk RNA-expression data for robust prognostic insights across cancer types

Background: Deconvolution of bulk RNA-expression data unlocks the cellular complexity of cancer, yet traditional pseudobulk benchmarks may not always be reliable in real-world settings where absolute cell proportions are unknown. Results: Here, we introduce a novel real-data framework, leveraging 18 real bulk RNA-expression cohorts (5,891 samples) across nine cancer types to evaluate five deconvolution methods based on differentially proportioned (DP) and prognosis-related (PR) cell types. Across three innovative benchmark scenarios—consistency with scRNA-seq, reproducibility across cohorts, and reproducibility of prognostic relevance—ReCIDE and BayesPrism stand out as two robust deconvolution methods. Application of a pan-cancer analysis based on the deconvolution of TCGA cohorts identifies matrix cancer-associated fibroblasts (mCAF) as a prognostic marker with consistent effects across multiple cancers. Building on this finding, we find a prognostic indicator combining classical monocytes and mCAF cell proportions to be significant in five TCGA cohorts, which we further validate in five independent GEO cohorts. Conclusions: This study broadens deconvolution benchmarking, offering actionable tools for precision oncology and guiding method selection for translational research.