Shuang Qin ¹ Haoxiang Zhang ² Chao Liu ³ Ming Yi ^4*

• ¹ Hubei Cancer Hospital, Wuhan, Hubei Province, China
• ² Shengli Clinical Medical College, Fujian Medical University, Fuzhou, Fujian Province, China
• ³ Shandong Cancer Hospital, Shandong University, Jinan, Shandong Province, China
• ⁴ Zhejiang University, Hangzhou, China

The treatment paradigm for lung cancer has evolved substantially over the past few decades [1]. Initially centered around traditional modalities such as radiotherapy and chemotherapy, the field has now shifted toward more sophisticated interventions, including targeted therapies and immunotherapy [2]. While these innovations have indeed improved outcomes and quality of life for many patients, a significant portion of individuals still do not experience substantial benefit from these advanced therapies [3]. In response, researchers have increasingly turned to multi-omics technologies to better understand the complex biology underlying lung cancer and its interactions within the immune microenvironment [4]. These high-throughput data, encompassing genomic, transcriptomic, proteomic, and metabolomic layers, offer rich information on potential therapeutic targets and prognostic markers that could redefine lung cancer treatment (Figure 1) [5,6]. However, the sheer scale and complexity of these datasets pose a critical interpretative challenge for clinicians and scientists alike.Enter deep learning, an advanced form of artificial intelligence (AI) that has proven instrumental in decoding large, complex datasets with unprecedented precision [7].Unlike traditional data analysis techniques, deep learning can parse intricate biological relationships and uncover novel oncogenic pathways that may otherwise remain obscured [8]. Through these capabilities, deep learning provides an invaluable lens through which researchers can gain insight into the dynamic processes driving lung cancer progression. It enables the in-depth analysis of complex phenomena such as cell signaling interactions, immune responses, and metabolic reprogramming within the tumor microenvironment [9]. As deep learning continues to advance, it holds the promise of not only improving our understanding of malignant biological behaviors but also driving the development of precision oncology strategies tailored to individual patients' unique disease profiles [10]. This special issue presents seven articles that investigate the role of deep learning in studying lung cancer, with a particular focus on the tumor immune microenvironment.These studies explore different ways deep learning is being applied to analyze the immune landscape, offering valuable perspectives on how AI can enhance our ability to predict therapeutic responses and identify new therapeutic targets.One such study, conducted by Zheng et al., examines the impact of STK11 mutations on patient outcomes in non-small cell lung cancer (NSCLC). In an analysis of 188 NSCLC patients, the researchers found that high STK11 expression correlates with improved progression-free and overall survival, an observation further substantiated by data from the TCGA cohort. However, when mutated, STK11 is associated with poorer outcomes in both lung squamous cell carcinoma and adenocarcinoma subtypes. To investigate these findings further, the team conducted bioinformatics analyses that revealed seven immune-related genes (CALCA, BMP6, S100P, THPO, CGA, PCSK1, and MUC5AC) that were overexpressed in STK11-mutated tumors. This overexpression suggests that STK11 mutation may drive specific changes in immune gene expression, which in turn can affect NSCLC prognosis. These data underscore the complex role of STK11 in lung cancer and demonstrate its potential as a target for more personalized therapeutic approaches.Another important study in this issue, led by Liu et al., explores the development of a novel AI-based immunoscore called the patho-immunoscore to predict outcomes in advanced non-squamous NSCLC patients undergoing chemoimmunotherapy. Using over 1,300 whole-slide images from the TCGA-LUAD dataset, the researchers built a model that demonstrated robust predictive performance, which was further validated across independent study cohorts, including CPTAC-LUAD and ORIENT-11. A high patho-immunoscore was associated with significantly improved progression-free survival in patients receiving chemoimmunotherapy, highlighting the potential of AIdriven immunoscoring as a powerful prognostic tool. Importantly, these results suggest that the patho-immunoscore may be a broadly applicable biomarker not only for NSCLC but also for other cancer types where immunotherapy plays a crucial role.The need for accurate biomarkers to predict responses to immunotherapy in NSCLC is further explored in an insightful review by Shi's team. They argue that conventional imaging approaches, which primarily capture macroscopic tumor changes, may fall short in meeting the precision required by modern cancer diagnosis and treatment. In contrast, CT and PET/CT radiomics can reveal molecular-level features, such as PD-1/PD-L1 expression and tumor mutation burden, that hold potential as indicators of immunotherapy efficacy and patient prognosis. By integrating radiomics with machine learning and AI, the researchers propose a novel diagnostic framework capable of assessing not only the therapeutic response but also the likelihood of immune-related side effects. This review positions radiomics as a promising non-invasive tool for predicting immunotherapy benefits in NSCLC, with the potential to facilitate more personalized treatment plans.