SMILES to ZINC Data in torch_geometric conversion function

[andreibrasoveanu97]

Hello,

I've trained some models on the ZINC dataset and I would like to use them to predict the values for molecules outside the dataset, given under a SMILES format. I've seen this discussion on this forum but I've spent hours reading through the links suggested there with no result.

I've found plenty of methods of converting SMILES to torch_geometric.Data objects, but not to the same format that is found within the torch_geometric.datasets.ZINC objects.

Does anyone know where could I find the function or a mapping of some kind to get the same result?

Thanks!

[rusty1s]

You can find the pre-processing code of ZINC here.

[andreibrasoveanu97]

This does not show however how are the molecules (or graphs) built, I am looking for the parsing method of the SMILES of a molecule to the actual graph (or Data object).

[rusty1s]

In this case I suggest to open an issue at https://github.com/graphdeeplearning/benchmarking-gnns