Extend PyG dataloader support about tdc

[kexinhuang12345]

hey, thanks for the suggestion! I believe we indeed have some basic node/edge attributes (atom/bond index, atom charge/chirality, etc) supported in the MolConvert class (

TDC/tdc/chem_utils.py

Line 3449 in afe08ca

def smiles2PyG(smiles):

). To clarify, are you saying we should provide more options for the atom/bond features?

[kexinhuang12345]

Note that we are also currently testing this new feature where you can also directly do something like this:

from tdc.single_pred import ADME
data = ADME(name = 'Caco2_Wang', convert_format = 'ECFP4')

It will create a separate column in the dataframe with the converted format.

However, as our current dataloader in default returns pandas dataframe, some of the convert format wouldn't work, for example, the DGL graph object would return error when we put it in a dataframe. So, in these cases, it would be ideal to simply return the DGL/PyG objects. We would work to complete this feature in the near future.