SHAP-interpretable machine learning integrating exposures and multi-omics reveals immune alterations and biomarkers in COPD-lung cancer comorbidity.

Introduction: Chronic obstructive pulmonary disease (COPD) and lung cancer (LC) frequently co-occur and share environmental and biological determinants, yet their cross-scale associations remain incompletely understood. Artificial intelligence and machine learning-based integration of exposome and multi-omics data provide new opportunities for dissecting this complex comorbidity. Methods: This study established a sequential, cross-scale framework integrating global epidemiological analysis of GBD data (1990-2021), exposome-wide risk factor assessment using random forest classification with SHAP interpretation, immune microenvironment characterization via CIBERSORT and single-cell RNA sequencing with LLM-assisted annotation, candidate gene prioritization through SMR analysis, LASSO regression, and machine learning classifiers, and experimental validation using RT-qPCR, western blotting, immunohistochemistry, dual immunofluorescence, and macrophage-epithelial Transwell co-culture. Results: COPD and LC demonstrated persistent global co-occurrence patterns across 204 countries and territories. SHAP analysis identified smoking, particulate matter pollution, and residential radon as major shared risk factors. Bulk and single-cell transcriptomic analyses revealed consistent immune microenvironment remodeling, with macrophages as the predominant shared immune population. Multi-omics intersection and machine learning prioritization identified TREM1 and ODF2L as shared hub genes, significantly downregulated in COPD and LC tissues. Macrophage-specific silencing of TREM1 or ODF2L attenuated macrophage-mediated promotion of A549 cell migration and proliferation, and dual immunofluorescence confirmed macrophage-associated localization of both proteins. Discussion: These findings provide a cross-scale perspective linking environmental exposures, macrophage-centered immune alterations, and COPD-LC comorbidity. TREM1 and ODF2L represent promising macrophage-associated candidate biomarkers and potential therapeutic targets. Integrating interpretable machine learning with exposome and multi-omics data offers a robust framework for biomarker discovery in chronic respiratory disease comorbidity.