Identification of Biomarkers for Sepsis-Induced Acute Lung Injury Through Bioinformatics and Machine Learning Approaches, with Experimental Validation

Yannian Luo,1,* Juan Xu,2,* Nannan He,1 Wen Cao1 1Department of Critical Care Medicine, The Second Hospital of Lanzhou University, Lanzhou, Gansu Province, People’s Republic of China; 2Department of Pediatric Neurology, The Second Hospital of Lanzhou University, Lanzhou, Gansu Province, People’s Republic of China*These authors contributed equally to this workCorrespondence: Wen Cao, Email [email protected]: Sepsis-induced acute lung injury (ALI) remains a life-threatening condition due to the lack of reliable early diagnostic biomarkers. Machine learning offers powerful tools for analyzing high-dimensional gene expression data and identifying potential biomarkers and therapeutic targets.Methods: Five datasets (GSE10474, GSE32707, GSE66890, GSE10361, GSE3037) were obtained from the GEO database. After assessment and normalization, GSE10474, GSE32707, and GSE66890 were combined as a training set to identify differentially expressed genes (DEGs). DEGs were intersected with genes from key modules identified by weighted gene co-expression network analysis (WGCNA), yielding 213 overlapping genes. These were analyzed via Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment. Eight machine learning algorithms (RF, SVM, GLM, GBM, KNN, NNET, LASSO, DT) were used to develop diagnostic models, which were validated on GSE10361 and GSE3037. Model performance was evaluated using a nomogram, calibration curves, and decision curve analysis (DCA). Immune and inflammatory states were assessed using the CIBERSORT algorithm. Potential therapeutic compounds were identified through the DSigDB database via the Enrichr platform. Molecular docking and molecular dynamics simulations examined interactions between Resveratrol and selected targets. In vitro experiments validated these findings.Results: A total of 213 candidate genes were identified by intersecting DEGs with WGCNA-derived MEblue module genes. GO and KEGG analyses indicated associations with immune activation and bacterial infection. Four key genes (DDAH2, PNPLA2, STXBP2, TCN1) were selected using eight machine learning algorithms. The diagnostic model showed good performance via nomogram, calibration curve, and DCA. Molecular docking revealed stable binding of Resveratrol to these genes. In vitro, Resveratrol pretreatment alleviated LPS-induced ALI by modulating the core genes.Conclusion: The four genes may serve as diagnostic biomarkers for sepsis-ALI. Resveratrol represents a potential therapeutic strategy by targeting these genes.Keywords: sepsis-induced ALI, machine learning algorithms, WGCNA analysis, CIBERSORT analysis, molecular docking