Graph deep-learning QSAR
Graph-based learning outperformed conventional models before and after augmentation. The GNN reached R2 0.98 and RMSE 0.02 after augmentation. The GNN performance is interpreted as evidence that molecular graph representations are suitable…
1 sources - 5 claims
Graph-based learning outperformed conventional models before and after augmentation. The GNN reached R2 0.98 and RMSE 0.02 after augmentation. The GNN performance is interpreted as evidence that molecular graph representations are suitable for xanthone-based QSAR. Graph neural networks represented molecules as graph-structured data with nodes and edges encoding molecular structure. Graph Attention Convolution dynamically weighted neighboring nodes to focus on structurally important neighborhoods.