Invalid SMILES not filtered by `filter_invalid_smiles`

[yunsiechung]

Hi,

In kermt/util/utils.py, the filter_invalid_smiles function should filter out invalid SMILES. However, when the SMILES string is invalid, mol = Chem.MolFromSmiles(datapoint.smiles) line will actually return None, and the next line, if mol.GetNumHeavyAtoms() == 0: fails because mol is None. This can be prevented by adding the following 3 lines:

def filter_invalid_smiles(data: MoleculeDataset) -> MoleculeDataset:
    """
    Filters out invalid SMILES.

    :param data: A MoleculeDataset.
    :return: A MoleculeDataset with only valid molecules.
    """
    datapoint_list = []
    for idx, datapoint in enumerate(data):
        if datapoint.smiles == '':
            print(f'invalid smiles {idx}: {datapoint.smiles}')
            continue
        mol = Chem.MolFromSmiles(datapoint.smiles)
        ######################## NEW LINES ########################
        if mol is None:
            print(f'invalid smiles parse {idx}: {datapoint.smiles}')
            continue
        ######################## NEW LINES ########################
        if mol.GetNumHeavyAtoms() == 0:
            print(f'invalid heavy {idx}')
            continue
        datapoint_list.append(datapoint)
    return MoleculeDataset(datapoint_list)

Some SMILES strings that were valid with a different version of RDKit were encountering error when I was running KERMT. Hope this can make the filtering function more robust.