Graph Attention Network-Based Prediction of Drug-Gene Interactions of Signal Transducer and Activator of Transcription (STAT) Receptor in Periodontal Regeneration

Introduction The signal transducer and activator of transcription-1 (STAT-1) are tightly controlled signaling pathways, with induced genes acting as positive and negative regulators. Persistent activation of the signal transducer and activator of transcription (STATs), particularly signal transducer and activator of transcription-3 (STAT-3) and signal transducer and activator of transcription-5 (STAT-5), is common in human tumors and cell lines. STAT molecules act as transcription factors, regulated by ligands like interferon-α (IFN-α), interferon-γ (IFN-γ), epidermal growth factor (EGF), platelet-derived growth factor (PDGF), interleukin-6 (IL-6) and interleukin-27 (IL-27). STAT-1 mutations can cause infections like periodontitis, a chronic inflammatory disease affecting gum tissue and bone. STAT-1 drug-gene interactions are being studied for therapeutic applications. Our study aims to predict drug-gene interactions of STAT-1 receptors in periodontal inflammation using graph attention networks (GATs). Methodology The study used a dataset of 215 drug-gene interactions to train and test a GAT model. The data was cleaned and normalized before being subjected to GATs using the Python library. Cytoscape and cytoHubba were used to visualize and analyze biological networks, including drug-gene interactome networks. The GAT model consisted of two graph attention layers, with the first layer producing eight features and the second layer aggregating outputs for binary classification. The model was trained using the Adam optimizer and CrossEntropyLoss function. Results The drug-gene interactome network, analyzed using Cytoscape, had 657 nodes, 1591 edges, and 4.755 neighbors. The predictive GAT model had low accuracy due to data availability and complexity. Conclusion The GAT model for drug-gene interactions in periodontal inflammation had low accuracy due to data limitations, complexity, and inability to capture all relevant features.