Multi-omics Reveals Metabolic-Inflammatory Drivers of Lung Cancer: An Integrated Mendelian Randomization and Machine Learning Study

Background: Metabolic dysregulation and chronic inflammation are implicated in the tumorigenesis of lung cancer; however, the causal relationships and translational potential of these associations remain poorly understood. This study aims to delineate the "metabolic-inflammatory-cancer" axis in order to identify high-confidence therapeutic targets and to develop a clinical prognostic tool. Methods: We employed a comprehensive systems biology framework that integrates univariate and multivariate analyses, bidirectional Mendelian Randomization (MR), Bayesian colocalization, and multi-omics validation, encompassing transcriptomics, proteomics, and methylomics. Utilizing extensive datasets from GWAS, TCGA, and CPTAC cohorts, we systematically evaluated 30 metabolic and inflammatory exposures. Subsequently, a machine learning-based LASSO-Cox regression model was developed to translate epigenetic findings into a prognostic risk score. Results: MR analysis revealed smoking initiation (\((OR=1.64--1.86)\)) and Body Mass Index (BMI) as predominant causal risk factors. Exploratory mediation analysis suggested a nominally significant role for C-reactive protein (CRP), which mediates approximately 20.15% of the BMI-driven risk for Lung Squamous Cell Carcinoma (LUSC) (nominal \((P < 0.05)\)). Integration of multi-omics data underscored the "thrombopoiesis/coagulation" pathway as a central pathogenic mechanism, corroborated by the upregulation of platelet-related signatures. We identified 51 high-confidence causal genes, which included CDK11A , a novel cell-cycle regulator, and MFAP2 . Notably, MFAP2 exhibited a dissociation between transcript and protein expression, suggesting complex post-transcriptional regulation within the tumor microenvironment. Clinically, our machine-learning derived DNA methylation risk score achieved a C-index of 0.782 within the training cohort, significantly surpassing traditional TNM staging. Importantly, the prognostic signature was successfully validated for Recurrence-Free Survival (RFS) in an independent external cohort (GSE39279, \((n=122)\)), where the continuous risk score was confirmed as a significant prognostic factor by univariate Cox regression (\((\text{HR} = 0.9972, P = 0.0142)\)). Although the association with Overall Survival (OS) was not significant (Log-rank \((P = 0.92)\)), the model demonstrated robust clinical utility for recurrence risk. Conclusions: This study elucidates the causal architecture underlying metabolic-driven lung cancer, highlighting the inflammation-coagulation axis as a pivotal therapeutic target. The developed methylation-based risk score provides a robust, actionable tool for precision recurrence risk stratification, effectively bridging the gap between genomic discoveries and clinical implementation.