Identifying Lipid Metabolism-Related Therapeutic Targets and Diagnostic Markers for Lung Adenocarcinoma by Mendelian Randomization and Machine Learning Analysis

BackgroundLipid metabolic disorders are emerging as a recognized influencing factors of lung adenocarcinoma (LUAD). This study aims to investigate the influence of lipid metabolism-related genes (LMRGs) on the diagnosis and treatment of LUAD and to identify significant biomarkers.MethodsDESeq2 and robust rank aggregation (RRA) analyses were employed to determine the differential expression of LMRGs from TCGA-LUAD and five GEO datasets. Mendelian randomization (MR) was conducted utilizing protein quantitative trait loci (pQTLs) in the deCODE, prot-a, and UKB-PPP Study to estimate causal relationships between plasma proteins and LUAD within the ieu-a-984, ieu-a-965, and FinnGen R10 cohorts as potential drug targets of LUAD. Subsequently, an optimal machine learning model for diagnosing LUAD was established by comparing four models: support vector machine, random forest (RF), glmBoost, and eXtreme Gradient Boosting. Finally, the diagnostic performance of five plasma proteins was validated through nomogram analysis, calibration curve assessment, decision curve analysis (DCA), independent internal and external datasets.ResultA total of five biomarkers were identified from 1034 LMRGs via MR and differential expression analysis. TNFRSF21 exhibited a positive association with LUAD risk; conversely, BCHE, FABP4, LPL, and PLBD1 demonstrated negative correlations with this risk. The RF machine learning model was determined to be the optimal model for diagnosing LUAD using these five plasma proteins. Ultimately, nomogram construction, calibration curve analysis, DCA, as well as independent internal and external dataset validation confirmed that these biomarkers exhibit excellent diagnostic performance.ConclusionsBCHE, FABP4, LPL, PLBD1, and TNFRSF21 represent potential novel reliable diagnostic markers as well as therapeutic targets for LUAD.