DiffDec is an end-to-end E(3)-equivariant diffusion model to optimize molecules through molecular scaffold decoration conditioned on the 3D protein pocket.

conda env create -f environment.yaml

Please refer to README.md in the data folder.

To train a model for single R-group decoration task, run:

python train_single.py --config configs/single.yml

To train a model for multi R-groups decoration task, run:

python train_multi.py --config configs/multi.yml

You can sample 100 decorated compounds for each input scaffold and protein pocket and change the corresponding parameters in the script. Run the following:

bash sample.sh

You will get .xyz and .sdf files of the decorated compounds in the directory sample_mols.

You can run evaluation scripts after sampling decorated molecules:

bash evaluate.sh