Global and local integrated gradient-based diffusion model for De novo drug design

Datasets

CrossDocked (TargetDiff)

BindingMOAD (GCDM)

ASBench

Data processing

Our backbone model is Targetdiff, so we followed the same procedure of TargetDiff.

Checkpoint

Pre-trained ClintDM

Code

Training: train_GlintDM.py (only trained on CrossDocked)

Sampling: sample_[CrossDocked/MOAD]_GlintDM.py

Evaluation: evaluate_[CrossDocked/MOAD]_results.py