Utilising Bioinformatics and Machine Learning for Identification of Neonatal Bronchopulmonary Dysplasia Characteristic Genes: Association with Immune Cell Infiltration and a Two-Sample Mendelian Randomisation Study of Immune Outcomes.

Background: : Bronchopulmonary Dysplasia is a relatively common disorder affecting preterm infants, impacting their health and future development. Although machine learning methodologies serve as valuable tools for making predictions and extracting biomedical insights, their use for in-depth analysis and prediction of diseases remains limited. In this study, we employed machine learning techniques to identify key genes associated with Bronchopulmonary Dysplasia and their correlation with immune cells. Method: : We screened a dataset of premature infants with Bronchopulmonary Dysplasia and downloaded the related datasets from the GEO database. Subsequently, we conducted an analysis of differentially expressed genes (DEGs). With Gene Ontology (GO) annotation, we classified DEGs according to the official classification and performed GO functional enrichment using the phyper function of R. KEGG pathway analysis was also conducted through pathway functional enrichment employing the same R function. We utilized Weighted Gene Co‐expression Network Analysis (WCGNA) to identify co‐expression gene modules, explore the relationship between the gene network and phenotype, and pinpoint key genes in the network. Genes were then fitted using the Least Absolute Shrinkage and Selection Operator (Lasso),Support Vector Machine (SVM) and Random Forest analysis. Following the characterization of differentially expressed genes, we employed Gene Set Pathway Enrichment analyses (GSEA) to identify enriched signaling pathways. Finally, the infiltration of immune cells was classified using CIBERSORT. Results: : Our analysis of GSE32472 identified a total of 273 differentially expressed genes (DEGs). When applying the WGCNA method to our dataset, we categorized 20002 gene expression traits into 21 modules. Remarkably, the red module showed a significant correlation with Bronchopulmonary Dysplasia (correlation=-0.6, p<0.0001). We utilized LASSO, SVM, and random forest algorithms to select signature genes, which included CCDC141, CHI3L2, PDLIM7, and RIMKLB. The receiver operating characteristic (ROC) curves for these signature genes were 0.928（95% CI: 0.761−0.969）, 0.880（95% CI: 0.803−0.971）, 0.900（95% CI: 0.841−0.986）, and 0.880（95% CI: 0.793−0.975）, respectively, and this was confirmed using an external dataset from GSE108754. Furthermore, to verify whether these selected genes are highly expressed in the mother and thus contribute to the newborn’s susceptibility to BPD, we used GSE188944 as a validation set. GSEA analysis suggested that these signature genes are involved in the spliceosome, sulfur metabolism, RNA degradation, aminoacyl-tRNA biosynthesis, T cell receptor signaling pathway, maturity onset diabetes of the young, PPAR signaling pathway, and glycosaminoglycan degradation, all of which are positively correlated. Conclusion: : CCDC141, CHI3L2, PDLIM7, and RIMKLB demonstrated significant diagnostic advantages for neonatal bronchopulmonary dysplasia and played a crucial role in immune cell infiltration.