Abstract
Introduction The Wnt/β-catenin pathway is crucial for bone formation and remodeling, regulating osteoblast differentiation, bone remodeling, and skeletal homeostasis. Dysregulation of the Wnt/β-catenin pathway is linked to bone-related diseases like osteoporosis, osteoarthritis, and osteosarcoma. The strategies to modulate this pathway include Wnt agonists, inhibitors, and small molecules. Graph neural networks (GNNs) have shown potential in understanding drug-gene interactions, providing accurate predictions, identifying novel drug-target pairs, and enabling personalized drug therapy. So we aim to predict GNN-based drug-gene interactions of Wnt/β-catenin pathway in bone formation. Methodology The drug-gene interactions of Wnt signaling were annotated and preprocessed using Cytoscape, a powerful tool for building drug-gene interactions. Data was imported, nodes representing drugs and genes were created, and edges represented their interactions. GNNs were used to prepare data for nodes, genes, and drugs. GNNs are designed to operate on graph-structured data, capable of learning complex relationships between the nodes. The architecture consists of several steps: graph representation, message passing, node representation update, graph-level readout, and prediction or output. A data representation system is a GNN with an Adam optimizer, 100 epochs, a learning rate of 0.001, and entropy loss. Results The network has 108 nodes, 134 edges, and 2.444 neighbors, with a diameter of 4, radius of 2, and characteristic path length of 2.635. It lacks clustering, sparse connectivity, wide connection variation, and moderate centralization. The GNN model's drug-gene interactions demonstrate high precision, recall, F1 score, and accuracy, with a high sensitivity to true-positives and low false-negatives. Conclusion The study employs a GNN model to predict drug-gene interactions in the Wnt/β-catenin pathway, demonstrating high precision and accuracy, but further research is needed.