-
Notifications
You must be signed in to change notification settings - Fork 1.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Feature Labels for MolGraphConvFeaturizer #3926
Comments
Hi @sumone-compbio , are you considering the following features as well? |
This sums up to 27 features, while there are 30 features. Please, let me know what I am missing. Thank you |
@VaishnaviMudaliar hi, is there any update? I really need to know the correct labels of these features for my thesis subsmission. |
@sumone-compbio Can you come by our office hours? (MWF at 9am PST) |
@rbharath hi, I hope I'm not late. So, all I need to know is what are the 30 default features of MolGraphConvFeaturizer. Following the source code, I'm not able to count all the 30 default features when I add them up. As you can see in my comment above I'm only able to add up 27 default features. I don't understand what I am missing following the source code. |
@sumone-compbio Try checking the source; where are you getting the 30 number? See in the code what is actually being put in. I don't have time right now to go through it, but I can show you where to look if you can join OH. |
@rbharath hi, it's mentioned in the source code that the default atom or node level features are 30 and for edge (bond) level features it's 11 by default. Also, when I'm running this featurizer on smiles it indeed returns 30 features for each atom. |
@rbharath one more thing I would like to add is the atom positions from the featurizer (by comparing the first 10 elements of each atom feature vector) are different from the atom index you would get using mol.GetAtomWithIdx() in rdkit. You can also verify this by comparing the positions of the atoms in the featurizer result with the function below (the function simply marks the atom index to the atom in the mol image): `def mol_with_atom_index(mol): mol = Chem.MolFromSmiles(smiles) This is a suggestion to keep the indices the same to avoid any result misinterpretation. E.g. I am using GNNExplainer, I wish to highlight the substructures contributing the most towards a certain prediction e.g. antibiotic or not, etc. If the indices from your featurizer differ from those from the rdkit mol object results can't be interpreted. Thank you |
@sumone-compbio Can you come by our office hours? I would be happy to discuss with you there |
@rbharath sure, thanks for the patience. I could also mail you if you want. |
❓ Questions & Help
Hi, I need help labeling the feature names of MolGraphConvFeaturizer. I went through the code and found the features. Please correct me where my count for the feature is wrong:
This sums up to 27 features, while there are 30 features. Please, let me know what I am missing. Thank you
The text was updated successfully, but these errors were encountered: