Neural Network-Based Prediction of Hub Genes in the Interactome of Hereditary Gingival Fibromatosis: An AI-Driven Bioinformatics Study

Abstract

Hereditary gingival fibromatosis (HGF) is a rare genetic disorder characterized by excessive gum growth, often presenting in childhood or adolescence. Symptoms include difficulties in speech, eating, oral hygiene, and psychological distress. Understanding the molecular mechanisms behind HGF is crucial for identifying potential therapeutic targets. This study aimed to predict interactomic hub genes in HGF using neural networks. We analyzed the GEO dataset GSE4250 using the geor2 tool to identify differentially expressed genes. Cytoscape and the CytoHubba plugin were employed to construct the interactome, ranking hub genes based on centrality scores. A neural network model with an 80:20 train-test split was used to predict hub and non-hub genes, achieving an AUC of 0.853, classification accuracy of 0.720, F1 score of 0.720, precision of 0.720, and recall of 0.720. The resulting network consisted of 147 nodes and 1092 edges, demonstrating moderate heterogeneity and connectivity. Ten key hub genes were identified, offering insights into the molecular basis of HGF. While the neural network model shows promising predictive capacity, further validation in larger cohorts is required. Adding predictive features and functional validation experiments could deepen understanding of HGF's biological mechanisms.