A commentary on “Integrative multi-omics identifies endoplasmic reticulum: stress-related prognostic biomarkers and CD99-mediated cell-cell communication in lung adenocarcinoma”

Dear Editor, We read with great interest the article by Rong and colleagues entitled “Integrative Multi-Omics Identifies Endoplasmic Reticulum Stress-Related Prognostic Biomarkers and CD99-Mediated Cell-Cell Communication in Lung Adenocarcinoma”[1], which has been published in the International Journal of Surgery. The authors skillfully integrated single-cell transcriptomics, spatial transcriptomics, and bulk RNA-seq to delineate endoplasmic reticulum stress (ERS) heterogeneity in lung adenocarcinoma (LUAD), identifying novel prognostic biomarkers and highlighting the role of CD99 in intercellular communication. This study represents a significant step forward in understanding ERS-driven tumor microenvironment remodeling. We commend the authors for their comprehensive approach and would like to offer several perspectives that may further strengthen the clinical applicability and mechanistic depth of their findings. This commentary adheres to the TITAN 2025 Guidelines for transparent reporting in biomedical research[2]. No artificial intelligence tools were used in its preparation. The writing of this work was completed without AI assistance. First, while the authors employed a robust machine learning framework combining ten algorithms to construct a prognostic model, the interpretability of the model remains limited. Integrating explainable artificial intelligence techniques, such as SHAP (Shapley Additive Explanations)[3], could elucidate the contribution of each ERS-related gene to risk prediction, enhancing clinical interpretability and facilitating the identification of key drivers. Additionally, multimodal deep learning architectures – such as transformer-based models for integrating histopathology images with transcriptomic data – could further capture cross-scale biological features and improve prognostic stability, as demonstrated in recent computational pathology studies[4]. Second, although the model was validated in two GEO cohorts, broader external validation across diverse clinical and multi-omics datasets is essential to ensure generalizability. Studies such as that by Liu et al have shown that multi-cohort validation reduces center-specific bias and enhances clinical reliability[5]. Furthermore, incorporating standardized pre-processing protocols for spatial and single-cell data – such as batch effect correction and staining normalization for spatial transcriptomics – would improve reproducibility across institutions[6]. Third, the identification of CD99 as a key mediator of ERS-related cell-cell communication is a novel and compelling finding. To deepen the mechanistic insight, we suggest integrating spatial proteomics (e.g., multiplex immunofluorescence) or metabolic imaging (e.g., MALDI-MSI) to validate the protein expression and functional state of CD99 and its interaction partners within the tumor niche. Such approaches could bridge computational predictions with tissue-level biology, as illustrated in recent immuno-metabolic studies of lung cancer[7]. Fourth, the study highlights ERS subtypes with distinct immune profiles and drug sensitivity. Expanding this framework to incorporate longitudinal clinical and omics data could enable dynamic risk prediction and therapy response monitoring. Leitner et al developed AUTO-Surv, an interpretable deep learning model that integrates multi-omics with clinical variables for survival prediction, offering a template for temporal modeling in LUAD[8]. Applying similar strategies to serial biopsy or liquid biopsy data may further personalize prognosis and treatment selection. Finally, while the pan-cancer analysis suggests broad relevance of ANLN and AURKB, functional studies in LUAD-specific models are needed to confirm their roles in ERS and therapy resistance. Organoid-based assays or in vivo perturbation experiments targeting CD99 and key ERS genes could provide direct evidence of their therapeutic potential and inform future drug development. In summary, Rong et al have provided a valuable multi-omics resource that advances our understanding of ERS in LUAD. Incorporating explainable AI, multi-center validation, spatial functional validation, and longitudinal modeling could further enhance the translational impact of this work, paving the way for more precise prognostic tools and targeted therapies.