ML for rdMol descriptors

[marcosfelt]

Thanks for this package! Is there a reason the code used the rdkit descriptors from the ML package vs rdMolDescriptors?

[github-actions[bot]]

🎉 Welcome to scikit-mol! 🧪✨ Thank you for opening your first issue! 🚀 Your feedback helps improve the project and makes a difference. 💡 If you have any questions or need guidance, don't hesitate to ask. We're here to help! 🤝

[EBjerrum]

No reason other than that was what was used when we implemented it. Is there a difference?

[marcosfelt]

I was wondering if there is a major difference. The two things I noticed in a quick look:

1. The ML package seems to have more descriptors than rdMolDescriptors. There is also pretty good overlap between the two, though not perfect.
2. I think that rdMolDescriptors is implemented in C++ while the ML package is all in python. I'm not sure if that has any impact on performance.

I searched the rdkit documentation and issues, and I couldn't find anything on this topic.

[marcosfelt]

I did a bit more digging, and I actually think both might rely on the same code?

https://github.com/rdkit/rdkit/blob/master/rdkit/ML/Descriptors/MoleculeDescriptors.py

[EBjerrum]

Yes, I think that's probably it. The rdMolDescriptors seems to be more focused on calls to singular dscriptors, where the ML has the configurable generator that can easily manage calculation of several different by giving it a list of descriptors. If you are interested in descriptors, I recently saw a new and fast package for descriptor calculation that I has not had the time to look at yet https://github.com/osmoai/osmordred. It could maybe be nice with a scikit-mol interface for that? let me know if you would like to contribute (hint-hint ;-).