🐈 Task: Fit Mordred on DrugBank and Impute values for empty columns

[DhanshreeA]

Summary

In general, this is what we want to do:

1. We want to use the community fork for mordred because it takes care of compatibility issues (in principle, still needs validation)
2. We will the Imputer on mordred values form the DrugBank data, and then just use that as the checkpoint for the hub.

Objective(s)

• Update the eos78ao repo to use mordredcommunity, and use latest numpy <2 and networkx versions
• "Fit" mordred on DrugBank, and save checkpoint
• Add the checkpoint to the model repo, and use that in the main.py
• Test the model

Documentation

DrugBank data available here.

For the Imputer, we can utilise this.

[DhanshreeA]

Just leaving notes for future:

• With mordredcommunity, the model can run within Python 3.9, and 3.10 environments as well, and not just 3.8 as it does right now.
• Pip freeze output for reproducibility:

mordredcommunity==2.0.6
networkx==3.2.1
numpy==1.26.4
packaging==24.2
pandas==1.3.5
pillow==11.1.0
python-dateutil==2.9.0.post0
pytz==2024.2
rdkit==2023.3.2
six==1.17.0
timeout-decorator==0.5.0

[GemmaTuron]

Hi @DhanshreeA !

We will pick this up with @arnaucoma24 so he has a better insight of how Ersilia Models are organised. Your report here is quite useful. Can you let me know if any code is pushed to the repo already? If it is not ready don't worry we will tackle it with Arnau