Decoding Epigenetic Enhancer–Promoter Interactions in Periodontitis via Transformer-GAN: A Deep Learning Framework for Inflammatory Gene Regulation and Biomarker Discovery

Abstract

Background

Widespread tissue destruction and dysregulated immune responses are hallmarks of periodontitis, a chronic inflammatory disease. Although enhancer–promoter (E–P) interactions play a crucial role in gene regulation, little is known about how they affect the epigenetic regulation of periodontal inflammation. By combining DNA methylation and gene expression data using a novel deep learning framework, this study sought to decode the E–P regulatory landscape in periodontitis.

Methods

We examined matched genome-wide DNA methylation (GSE173081) and RNA-seq (GSE173078) datasets with integrated features such as methylation differences, gene expression changes, correlation metrics and genomic distances. A Transformer-GAN forecasted functional E–P interactions by training as a binary classifier to differentiate positive and negative enhancer–promoter pairs. AUC-ROC and AUC-PRC scores were used to benchmark the model’s performance, while functional enrichment and network topology analyses were employed to validate its biological relevance.

Results

The Transformer-GAN model outperformed traditional methods, exhibiting strong predictive performance (AUC-ROC = 0.725, AUC-PRC = 0.723). With a mean correlation of 0.62 and a median genomic distance of 45.2 kb, we found 262 significant E–P interactions involving 134 enhancers and 186 target genes. Multiple enhancers controlled central inflammatory genes, such as IL-1β, IL-6, IL-8 and TNF, creating network hubs enriched in immune pathways, including TNF, NF-κB and IL-17 signalling. Strong correlations were found between enhancer hypomethylation, active histone marks and gene upregulation through integrative multi-omics analysis. Interestingly, E–P interaction scores outperformed clinical indices or gene expression in terms of predicting treatment response (F1-score: 0.82). The diagnostic accuracy of the five CpG biomarkers ranged from 85% to 90%.

Conclusion

Our integrative Transformer-GAN approach reveals a complex enhancer–promoter regulatory network underlying inflammatory gene expression in periodontitis. These results reveal new biomarkers and potential treatment targets while highlighting the significance of epigenetic regulation in disease pathogenesis.