Role of Residual Graph Attention Networks in Predicting Gene-Drug Associations for Therapeutics in Vascularized Oral Cyst and Tumours

Abstract

ABSTRACT

Introduction: Ameloblastoma and dentigerous cysts are aggressive odontogenic lesions that can cause tooth displacement or resorption. Vascular Endothelial Growth Factor (VEGF) is crucial for tumor growth and cyst expansion, with excessive angiogenesis leading to tumor growth. Gene mutations and stem cell markers SOX2 and OCT4 play roles in these lesions. The study uses Residual Graph Attention Networks (RGAN) to predict drug-gene angiogenic associations in ameloblastomas and dentigerous cysts. This helps understand growth patterns and recurrence risks, enabling targeted therapies to inhibit VEGF signaling pathways. The research promotes personalized medicine and multifaceted approaches to managing odontogenic lesions, aiming to predict gene-drug associations for therapeutics in vascularized oral cysts and tumors.

Methods: The study analyzed data on VEGF-associated drugs and genes using probe and drug sites, network analysis, and graph neural networks. Data was sourced from PubChem, ChEMBL, DrugBank, and gene databases like Ensembl and KEGG. A graph representing drugs and genes was constructed, with each drug and gene as a unique vertex. Network analysis used metrics like degree centrality, closeness centrality, and betweenness centrality to identify key nodes. Graph Neural Networks (GNNs) were used to analyze associations between drugs and genes. Cytoscape was used to import files and assign drugs and genes to analyze the interactome. At the same time, the CytoHubba plugin identified the top 20 hub drugs and genes using the maximum clique centrality method. It subjected them to residual attention graph neural networks for predictive modeling.

Results: The network, with 242 nodes and 333 edges, has a sparse structure with low clustering and density, moderate centralization, and an efficient computational process. The model's performance was assessed using various metrics, including accuracy, confusion matrix, precision, recall, F1-score, ROC-AUC, precision-recall curve, prediction confidence distribution, learning curves, and threshold-based accuracy. It demonstrated exceptional performance, with the best test accuracy of 97.53%, high precision, recall, and F1 scores.

Conclusion: The study of drug-gene associations, particularly the role of Vascular Endothelial Growth Factor (VEGF) in ameloblastoma and dentigerous cysts, has significant implications for understanding these conditions and refining clinical management strategies. Ameloblastomas are aggressive odontogenic tumors with a propensity for recurrence, making their management challenging. Targeting the VEGF signaling pathway could limit tumor growth and invasive tendencies, modifying the treatment landscape for ameloblastoma patients. The TARGETS framework has introduced a predictive model for tailored treatments based on unique tumor biology and genetic makeup.