Wiesweg M, Mairinger F, Reis H, et al. - In non-small cell lung cancer (NSCLC), characteristics of tumor cells are primarily the focus of current predictive biomarkers for PD-1 (programmed cell death protein 1)/PD-L1 (programmed death-ligand 1)-directed immunotherapy, though major roles are played by tumor microenvironment and immune context in governing therapy response. Consequently, in diagnostic biopsies of patients with stage IV NSCLC, researchers sought to apply context-sensitive feature selection and machine learning strategies on expression profiles of immune-related genes. The NanoString nCounter platform was used to determine RNA expression levels in formalin-fixed paraffin-embedded tumor biopsies retrieved during the diagnostic workup of stage IV NSCLC from two thoracic oncology centers. They used a 770-gene panel covering immune-related genes and control genes. Supervised machine learning methods were applied for feature selection and generation of predictive models. Based on a training cohort of 55 patients with recurrent NSCLC treated with PD-1/PD-L1 antibody therapy, they selected features and created the model. They were able to identify patients with superior outcomes to immunotherapy with the resulting models, which were verified in two subsequently recruited, separate patient cohorts. Patients with highly favorable outcomes were identified by selecting by PD-L1 positivity at immunohistochemistry plus model prediction, and multivariate analysis confirmed the independence of PD-L1 positivity and model predictions. Visualization of the models revealed the predictive superiority of the entire 7-gene context over any single gene. By integrating tumor immune context, machine-learning approach enhances precision. Relative to single markers, gene expression context is identified to be more powerful.