Comparing gradient boosting and neural networks in the prediction of intersecting genes in gingival epithelial immunity.

Background: The gingival epithelium serves as the primary immune barrier against microbial invasion. Disruption contributes to chronic inflammation and the development of periodontitis. Understanding the gene interactions that regulate epithelial immunity is vital for effective diagnostics and treatment. Machine learning predicts key genes from differential and co-expression analyses, providing new insights into immune regulation. This study compares gradient boosting (GB) and neural network (NN) models in predicting genes involved in gingival epithelial immunity. Materials and Methods: Differential gene expression was analyzed using the iDEP web tool on the NCBI GEO dataset (GSE243173), which included samples from healthy, periodontitis, and LAD1 periodontitis subjects. Weighted gene co-expression network analysis (WGCNA) identified gene modules that were correlated with specific phenotypes. Intersecting genes from differential expression and WGCNA were preprocessed and classified using GB and NN models in the Orange platform. Model performance was evaluated using accuracy, precision, recall, specificity, F1 score, and AUC metrics to assess the predictive efficacy of gingival epithelial immune gene clusters. Results: The NN model outperformed GB in predicting clustered genes, accurately identifying positive and negative samples. It achieved an area under the curve (AUC) of 0.987, a classification accuracy of 0.958, an F1 score of 0.958, a precision of 0.963, a recall of 0.958, and a specificity of 0.979. These results demonstrate the potential of NN in identifying intersecting genes involved in gingival epithelial immunity. Conclusions: The NN model's superior performance suggests its potential in genomic studies, particularly in identifying genes involved in immune responses. Further optimization and validation are necessary to fully explore its capabilities.