search

thomasly / SubGNN

Subgraph Neural Networks (NeurIPS 2020)
https://zitniklab.hms.harvard.edu/projects/SubGNN
MIT License
1 stars 0 forks source link

SubGNN for chemicals #5

Open thomasly opened 3 years ago

thomasly commented 3 years ago

Convert SubGNN model (SuGNN/SubGNN.py, class SubGNN(pl.LightningModule)) to work on chemicals. This is a general issue. We can discuss the details here and open more specific issues as needed.

thomasly commented 3 years ago

Instead of modifying the SubGNN model, we can build a graph-level classification model on top of it. To do that, we need to generate subgraphs for each chemical and precompute graph metrics with their precompute_graph_metrics.py script. Then the subgraphs of chemicals can be fed to the graph-level model with mini-batches.

In addition, the node embeddings used in the original work (generated with node2vec) can be replaced by atom features or atom features + node2vec.