Integration of Single-Cell and Bulk RNA Sequencing Data using Ecotype Machine Learning for Prognostic Biomarker Discovery in Gastric Cancer

Background EcoTyper is a new machine learning framework, this work attempted to constructed an EcoTyper-related prognostic model for gastric cancer (GC). Methods The scRNA-seq data and bulk RNA-seq data for GC were obtained from the GEO and TCGA databases, respectively. Cell composition deconvolution was performed using CIBERSORTx. EcoTyper was employed for de novo discovery of scRNA-seq cell states and communities. Weighted Correlation Network Analysis was applied to explore the gene co-expression networks in GC. Subsequently, a risk model for ecotypes was constructed using bulk RNA-seq data. Results This work revealed the significant differences in cell distribution between normal and primary samples. Primary tumor samples showed a predominant presence of immune cells, including monocytes/macrophages and neutrophils. These immune cells were classified into two EcoTypers, E1 and E2, with E2 closely linked to primary tumor samples. Using ecotype-related risk scores, GC patients were stratified into high-risk (HR) and low-risk (LR) groups. HR patients exhibited worse overall survival and heightened sensitivity to Mirin, Oxaliplatin, Ruxolitinib, VE-822, and MG-132. Notably, the core gene TGM2 was up-regulated in GC cells, and its silencing reduced GC cell proliferation, migration, and invasion. Conclusion This study constructed a meaningful EcoTyper prognostic model, which served as a potential prognostic biomarker for GC treatment. This prognostic model showed significant correlations with immunotherapy and chemotherapy. This research has provided a potential valuable target for GC treatment.