Patients with early-stage lung adenocarcinoma (LUAD) exhibit different overall survival (OS) rates and immunotherapy responses. Understanding the immune landscape facilitates the personalized treatment of LUAD. The immune cell populations in tumour tissues were quantified to depict the immune landscape in early-stage LUAD patients in The Cancer Genome Atlas (TCGA). Early-stage LUAD patients in three immune clusters identified by the immune landscape exhibited different survival potentials. A prognostic immune-related gene signature was built to predict the survival of early-stage LUAD patients. Several machine learning methods (support vector machine, naive Bayes, random forest, and neural network-based deep learning) were applied to train the classifiers to identify the immune clusters in early-stage LUAD based on the gene signature. The four classifiers exhibited a robust effect in identifying the immune clusters. A random forest regression model identified that TP53 was the most important gene mutation associated with the immune-related signature. Furthermore, a decision tree and a nomogram were constructed based on the immune-related gene signature and clinicopathological traits to improve risk stratification and quantify risk assessment for individual patients. Five external test cohorts were applied to validate the accuracy of the immune-related signature. Our study might contribute to the development of immunotherapy and the personalized treatment of early-stage LUAD. KEY MESSAGES: Immune landscape correlates with the clinical outcome of early-stage adenocarcinoma (LUAD). Machine learning methods identifies a prognostic gene signature to predict the survival and prognosis of early-stage LUAD. TP53 gene mutation status correlates with the immune landscape in early-stage LUAD.
Keywords: Early-stage lung adenocarcinoma (LUAD); Immune landscape; Overall survival; Prognostic model.