e14664 Background: Checkpoint inhibitor therapy has become a cornerstone in the treatment of non-small cell lung cancer (NSCLC). However, the response to checkpoint inhibitors varies widely among patients, with some experiencing significant benefits and others facing adverse reactions without therapeutic gain. The ability to accurately predict which patients will benefit from checkpoint inhibitor therapies is essential for enhancing treatment efficacy and minimizing side effects. In this context, the identification and analysis of predictive biomarkers, including the expression of immune checkpoint proteins such as PD-L1 and TIGIT, and the comprehensive profiling of cytokines in the tumor microenvironment, are of paramount importance. Methods: A multimodal strategy was employed to identify predictive biomarkers for checkpoint inhibitor therapy in non-small cell lung cancer (NSCLC). This approach integrated metabolic, transcriptomic, and proteomic data to achieve a thorough insight into the biomarkers likely to influence patient responses. The diverse data sets were fused through advanced machine learning techniques, creating a holistic dataset. Analysis of this integrated dataset facilitated the discovery of multimodal biomarker signatures, offering the potential to predict which NSCLC patients would positively respond to checkpoint inhibitor treatments. Results: Initial observations revealed considerable variability in the expression of PD-L1 and TIGIT alongside other immune checkpoint molecules, across different regions of the tumor microenvironment in non-small cell lung cancer (NSCLC). Multimodal biomarker signatures, encompassing distinct expressions of immune checkpoint inhibitors, cytokine levels, and the presence of immune cells, showed a strong correlation with patient outcomes following treatment with immune checkpoint inhibitors. Conclusions: Our research highlights the significant promise of using a multimodal stratification method to unravel the complexities of non-small cell lung cancer (NSCLC) and to predict patient responses to immune checkpoint inhibitors. Through the integration of metabolomic, transcriptomic, and proteomic data, we discovered distinctive biomarker signatures that were highly predictive of responses to these immunotherapies.