Integrative Transcriptomic and Machine Learning Analysis Identifies JAG1 and NR2F2 in Keloid.

OBJECTIVE: To identify key genes associated with keloid through integrative analysis of multiple transcriptomic datasets, and to explore their potential regulatory roles in disease development with experimental validation. METHODS: Transcriptomic data of keloid were retrieved from the Gene Expression Omnibus (GEO) database. Batch effects were corrected using SVA and ComBat. Differentially expressed genes (DEGs) were identified using the limma package. Weighted gene co-expression network analysis (WGCNA) was performed to detect key modules, and intersected with DEGs to obtain candidate genes. Multiple machine learning algorithms were applied to identify key feature genes, which were further validated in independent external datasets. In parallel, wound tissues from a silicone-splinted murine excisional wound model (which recapitulates mechanical tension of human skin injury) and matched normal skin tissues were collected for histological assessment using haematoxylin-eosin (H&E) and Masson staining, and gene expression levels were validated by quantitative real-time PCR (qRT-PCR). RESULTS: A total of 82 DEGs were identified. WGCNA revealed the MEpink module as the key module, and intersection analysis yielded 61 candidate genes, which were primarily enriched in inflammation- and fibrosis-related pathways. Protein-protein interaction network analysis indicated that JAG1 and NR2F2 occupied central positions within the network. Machine learning identified six key feature genes, among which NR2F2 and IL15 showed relatively strong discriminatory performance. Histological analysis demonstrated increased collagen deposition and disorganized tissue architecture in keloid samples. qRT-PCR confirmed that JAG1 and NR2F2 were upregulated in keloid tissues, consistent with transcriptomic findings. CONCLUSION: JAG1 and NR2F2 participate in inflammation- and fibrosis-related signaling cascades to drive keloid pathogenesis, and possess promising value for the molecular subtyping of keloid lesions. Collectively, our results lay a solid foundation for future mechanistic research on the molecular regulatory networks governing keloid development.