Leveraging LSGAN for synthetic gingival keratinization genomic data: Insights from drug interaction and gene ontology in an early fusion framework

Introduction: Gingival keratinization, a vital process in oral health, involves the formation of a keratin-rich protective epithelial layer, providing resilience against mechanical stress, pathogens, and environmental factors. Objective: This study employs an early fusion omics approach with Least-Squares Generative Adversarial Networks (LSGAN) to generate synthetic genomic data, incorporating insights from drug interactions and geneontology annotations. Methods: Gene expression data from the NCBI GEO dataset (GSE182196) were analyzed to identify differentially expressed genes (DEGs) across diverse samples. Functional enrichment was performed using the Comparative Toxicogenomics Database (CTD) to explore chemical exposures and biological processes linked to DEGs. Outputs were standardized into TSV formats for downstream analyses. To ensure high-fidelity synthetic data generation, the LSGAN framework was optimized to minimize Mean Squared Error (MSE) and Mean Absolute Error (MAE).