Machine learning-based risk model incorporating tumor immune and stromal contexture predicts cancer prognosis and immunotherapy efficacy

Abstract

Summary: The immune and stromal contexture within the tumor microenvironment (TME) interact with cancer cells and jointly determine disease process and therapeutic response. We aimed at developing a risk scoring model based on TME-related genes of squamous cell lung cancer to predict patient prognosis and immunotherapeutic response. TME-related genes were identified through exploring genes that correlated with immune scores and stromal scores. LASSO-Cox regression model was used to establish the TME-related risk scoring (TMErisk) model. A TMErisk model containing six genes was established. High TMErisk correlated with unfavorable OS in LUSC patients and this association was validated in multiple NSCLC datasets. Genes involved in pathways associated with immunosuppressive microenvironment were enriched in the high TMErisk group. Tumors with high TMErisk showed elevated infiltration of immunosuppressive cells. High TMErisk predicted worse immunotherapeutic response and prognosis across multiple carcinomas. TMErisk model could serve as a robust biomarker for predicting OS and immunotherapeutic response.

Keywords

• Biological sciences
• Immunology
• Cancer
• Machine learning