reading notes of《Molecular Graph Representation Learning and Generation for Drug Discovery》
文章目录
- The typical GNN architecture that relies on local aggregation operations can often miss higher-order graph information. In the context of molecular problems, locality can prove myopic, as longer-range dependencies are important for many biological prediction targets. To remedy this, we propose Path-Augmented Graph Transformer Networks (PAGTN) that are explicitly built on longer-range dependencies in graph-structured data.
- For large enough graphs, GCNs may fail to capture these long-range dependencies entirely. Our PAGTN model can more easily capture these dependencies because every node attends to all other nodes in the graph.
- GCN models necessitate several layers to propagate information and learn these substructures. The advantage of our model is that this interaction can be learned within a single layer.