Multi-omics integration and machine learning identify a novel gene signature consisting of CCDC141, CHI3L2, RIMKLB, and PDLIM7 in bronchopulmonary dysplasia associated with neutrophil-driven immune dysregulation.

BACKGROUND

Bronchopulmonary Dysplasia (BPD) is a major complication in preterm infants, lacking effective diagnostic biomarkers to facilitate early detection and intervention.

METHODS

We employed a multi-omics approach to identify diagnostic biomarkers for BPD. Analysis of the GEO dataset GSE32472 revealed 273 Differentially Expressed Genes (DEGs). Weighted Gene Co-expression Network Analysis (WGCNA) identified co-expression modules. A diagnostic model was developed using a machine learning pipeline that sequentially integrated LASSO regression, Support Vector Machine (SVM), and random forest algorithms. Feature selection through machine learning identified a four-gene diagnostic signature. We conducted Gene Set Enrichment Analysis (GSEA) and immune cell infiltration assessment via CIBERSORT. Additionally, a two-sample Mendelian Randomization (MR) analysis explored the potential causal relationship between serum neutrophil counts and BPD-related lung impairment. Experimental validation employed a BPD animal model with H&E staining, TUNEL assay, RT-qPCR, and Western blotting.

RESULTS

WGCNA identified 21 co-expression modules, with the red module showing the strongest correlation with BPD. Machine learning feature selection identified a four-gene diagnostic signature: CCDC141, CHI3L2, PDLIM7, and RIMKLB. This signature demonstrated significant diagnostic potential (AUC > 0.8), validated in an external dataset. GSEA indicated associations with immune and metabolic pathways. Furthermore, MR analysis suggested a causal link between higher serum neutrophil counts and reduced risk of BPD-related lung impairment.

CONCLUSIONS

CCDC141, CHI3L2, PDLIM7, and RIMKLB are identified as promising diagnostic biomarkers for BPD ‒ potentially involved in immune-related pathogenesis.