Abstract
BACKGROUND: Immunotherapy benefits shows discrepancy in different lung adenocarcinoma (LUAD) patients because of the intratumor heterogeneity (ITH). ITH favors tumor evolution and correlated with drug resistance. The genes mediating ITH in LUAD and their role in predicting prognosis and therapy benefits is unclear. METHODS: An ITH-related signature (IRS) was built by ten methods-based integrative machine learning programs using TCGA, GSE68571, GSE42127, GSE30129, GSE50081, GSE72094, GSE37745, GSE68467, and GSE31210 dataset. To assess the relationship between IRS and the tumor immune microenvironment, numerous prediction scores were employed. RESULTS: The optimal predictive signature for LUAD cases was the IRS developed using Lasso + stepCox(both) method, which had the highest average C-index of 0.80. It performed consistently and effectively in predicting the clinical outcomes of LUAD patients. Additionally, compared to the clinical stage and numerous other existing prediction models, a higher C-index was demonstrated in IRS. LUAD patients with low IRS score had a higher level of immuno-activated cells, higher TMB score, lower ITH score, higher PD1&CTLA4 immunophenoscore, and tumor escape score in LUAD. The gene set score for angiogenesis, coagulation, hypoxia, and NOTCH signaling were increased in LUAD with high IRS score. CONCLUSION: Overall, the study developed a unique IRS for LUAD that may serve as a predictor of the clinical outcome and immunotherapy advantages for individuals with LAUD.