"NMF Deconvolution of a High-ROS Transcriptional Program Uncovers mTOR-Dependent Therapeutic Sensitivity in Stomach Adenocarcinom"

"Stomach adenocarcinoma is driven by heterogeneity,limiting therapeutic success. Although ROS acts as a continuous\u201credox rheostat\u201d for tumor evolution, it is categorized basedon binary models that are masked by tumor-microenvironment(TME) confounders. Here, we have defined a continuous, TME-independent ROS axis to help identify intrinsic vulnerabilitiesand improve patient stratification. We have used non-negativematrix factorization (NMF) to define a ROS-Axis in TCGA-STAD which was validated in ACRG Cohort. Using multivariateregression model we isolated intrinsic signatures via \u201cresidual\u201dROS scores by adjusting for TME confounders. Survival was as-sessed using Cox hazard models. Drug sensitivities were mappedusing GDSC2\/ElasticNet modeling with cross-cohort replication.Our results define a reproducible ROS gradient, driven byeffectors like NQO1 and SOD1, characterizing ROS-high tumorsas proliferative, epithelial and \u201cimmune-cold\u201d. High residualROS score was associated with an improved prognosis, regardlessof TNM stage and age. Pharmacogenomic mapping revealed anoverlapping sensitivity to mTOR inhibitors in ROS-high gastriccancer tumors which persisted after TME confounder adjust-ment. The continuous ROS axis provides a functional readout ofmetabolic dependency that refines traditional anatomical staging.By identifying mTOR dependent cold tumors, our frameworkoffers a precision strategy for immunotherapy-resistant patientslike those affected by microsatellite-stable gastric cancer."