Interpretable artificial intelligence based on immunoregulation-related genes predicts prognosis and immunotherapy response in lung adenocarcinoma.

Introduction: Lung adenocarcinoma (LUAD) is the most common subtype of non-small cell lung cancer, and its benefit from immune checkpoint inhibitors (ICIs) is controversial, especially for patients without driver gene mutations. The potential of immunoregulation-related genes (IRGs) in predicting the prognosis of LUAD and the efficacy of immunotherapy becomes emerging. There is an urgent need to establish a reliable IRGs-based predictive model of ICI response.

Methods: Extract and merge LUAD RNA sequencing data and clinical data from GEO database. The differences in genomic and tumor microenvironment (TME) cell infiltration landscape between normal lung tissue and tumor tissue were comprehensively analyzed. Unsupervised consistent cluster analysis based on genes related to immune regulation was performed on the samples. ESTIMATE and TIMER algorithms were used to analyze the infiltration of immune cells in different groups, and TIDE score was used to evaluate the effectiveness of immunotherapy. Then, lasso regression was used to establish a prognostic model based on identified key IRGs. XGBoost machine learning algorithm was further developed, with SHapley Additive exPlanations (SHAP) to interpret the model.

Results: The GEO LUAD cohort was divided into two clusters based on IRG expression, with significantly better survival outcomes and immune cell infiltration in the IRG-high group compared to the IRG-low group. TIDE scores indicated that the group with high IRG pattern showed a better response to ICI treatment. Then, we developed an IRG index (IRGI) model based on identified 2 key IRGs, GREM1 and PLAU, and IRGI effectively divided patients into high-risk and low-risk groups, revealing significant differences in prognosis, mutational profiles, and immune cell infiltration in the TME between two groups. Subsequently, the interpretable XBGoost machine learning model established based on IRGs could further improve the predictive performance (AUC = 0.975), and SHAP analysis demonstrated that GREM1 had the greatest impact on the overall prediction.

Discussion: IRGI can be used as a valuable biomarker to predict LUAD patient prognosis and response to ICIs. IRGs play a crucial role in shaping the diversity and complexity of TME cell infiltration, which may provide valuable guidance for ICI treatment decisions for LUAD patients.