Thesis: Integration Chemical Knowledge into GNN

[aditya0by0]

Primary Tasks:

• Rule-based Functional Group detection including rings
• Augmented GNN with artificial nodes for each Functional group
• GAT Message passing mechanism

[aditya0by0]

@sfluegel05 ,
I was reviewing the GraphPropertyReader class and I have a doubt in the _read_data method, the graph seems to be constructed as directed, since the edge_index is populated only in one direction (i.e., [begin_atom_idx] -> [end_atom_idx]).

edge_index = torch.tensor(
    [
        [bond.GetBeginAtomIdx() for bond in mol.GetBonds()],
        [bond.GetEndAtomIdx() for bond in mol.GetBonds()],
    ]
)

Just want to make sure that — is this an intentional decision to work with directed graphs for molecules? Or is this converted to an undirected graph later in the pipeline which I might have missed?

Since chemical bonds are usually considered undirected in molecular graph models, so I had this doubt.

[aditya0by0]

@sfluegel05, I'm currently encountering another issue while running the graph repository. Are you experiencing the same error while running it from your end as well?

G:\anaconda3\envs\gnn\python.exe G:\github-aditya0by0\python-chebai\chebai fit --trainer=configs/training/default_trainer.yml --trainer.callbacks=configs/training/default_callbacks.yml --model=G:/github-aditya0by0/python-chebai-graph/configs/model/gnn_res_gated.yml --model.train_metrics=configs/metrics/micro-macro-f1.yml --model.test_metrics=configs/metrics/micro-macro-f1.yml --model.val_metrics=configs/metrics/micro-macro-f1.yml --data=G:/github-aditya0by0/python-chebai-graph/configs/data/chebi50_graph.yml --model.criterion=configs/loss/bce.yml --data.init_args.batch_size=40 --data.init_args.num_workers=12 --model.pass_loss_kwargs=false --data.init_args.chebi_version=227 --trainer.min_epochs=1 --trainer.max_epochs=1 

error: Parser key "data":
  Does not validate against any of the Union subtypes
  Subtypes: [<class 'NoneType'>, <class 'lightning.pytorch.core.datamodule.LightningDataModule'>]
  Errors:
    - Expected a <class 'NoneType'>
    - metaclass conflict: the metaclass of a derived class must be a (non-strict) subclass of the metaclasses of all its bases
  Given value type: <class 'dict'>
  Given value: {'class_path': 'chebai_graph.preprocessing.datasets.chebi.ChEBI50GraphData'}