DecompDiff: Diffusion Models with Decomposed Priors for Structure-Based Drug Design

This repository is the official implementation of DecompDiff: Diffusion Models with Decomposed Priors for Structure-Based Drug Design.

Dependencies

Install via Conda and Pip

conda create -n decompdiff python=3.8
conda activate decompdiff
conda install pytorch pytorch-cuda=11.6 -c pytorch -c nvidia
conda install pyg -c pyg
conda install rdkit openbabel tensorboard pyyaml easydict python-lmdb -c conda-forge

# For decomposition
conda install -c conda-forge mdtraj
pip install alphaspace2

# For Vina Docking
pip install meeko==0.1.dev3 scipy pdb2pqr vina==1.2.2 
python -m pip install git+https://github.com/Valdes-Tresanco-MS/AutoDockTools_py3

Preprocess

python scripts/data/preparation/preprocess_subcomplex.py configs/preprocessing/crossdocked.yml

We have provided the processed dataset file here.

Training

To train the model from scratch, you need to download the *.lmdb, *_name2id.pt and split_by_name.pt files and put them in the data directory. Then, you can run the following command:

python scripts/train_diffusion_decomp.py configs/training.yml

Sampling

To sample molecules given protein pockets in the test set, you need to download test_index.pkl and test_set.zip files, unzip it and put them in the data directory. Then, you can run the following command:

python scripts/sample_diffusion_decomp.py configs/sampling_drift.yml  \
  --outdir $SAMPLE_OUT_DIR -i $DATA_ID --prior_mode {ref_prior, beta_prior}

We have provided the trained model checkpoint here.

If you want to sample molecules with beta priors, you also need to download files in this directory.

Evaluation

python scripts/evaluate_mol_from_meta_full.py $SAMPLE_OUT_DIR \
  --docking_mode {none, vina_score, vina_full} \
  --aggregate_meta True --result_path $EVAL_OUT_DIR

Results

• JSD of bond distances

Bond	liGAN	GraphBP	AR	Pocket2Mol	TargetDiff	Ours
C-C	0.601	0.368	0.609	0.496	0.369	0.359
C=C	0.665	0.530	0.620	0.561	0.505	0.537
C-N	0.634	0.456	0.474	0.416	0.363	0.344
C=N	0.749	0.693	0.635	0.629	0.550	0.584
C-O	0.656	0.467	0.492	0.454	0.421	0.376
C=O	0.661	0.471	0.558	0.516	0.461	0.374
C:C	0.497	0.407	0.451	0.416	0.263	0.251
C:N	0.638	0.689	0.552	0.487	0.235	0.269

• JSD of bond angles

Angle	liGAN	GraphBP	AR	Pocket2Mol	TargetDiff	Ours
CCC	0.598	0.424	0.340	0.323	0.328	0.314
CCO	0.637	0.354	0.442	0.401	0.385	0.324
CNC	0.604	0.469	0.419	0.237	0.367	0.297
OPO	0.512	0.684	0.367	0.274	0.303	0.217
NCC	0.621	0.372	0.392	0.351	0.354	0.294
CC=O	0.636	0.377	0.476	0.353	0.356	0.259
COC	0.606	0.482	0.459	0.317	0.389	0.339

• Main results

Methods	Vina Score (↓)	Vina Min (↓)	Vina Dock (↓)	High Affinity (↑)	QED (↑)	SA (↑)	Success Rate (↑)
Reference	-6.46	-6.49	-7.26	-	0.47	0.74	25.0%
liGAN	-	-	-6.20	11.1%	0.39	0.57	3.9%
GraphBP	-	-	-4.70	6.7%	0.45	0.48	0.1%
AR	-5.64	-5.88	-6.62	31.0%	0.50	0.63	7.1%
Pocket2Mol	-4.70	-5.82	-6.79	51.0%	0.57	0.75	24.4%
TargetDiff	-6.30	-6.83	-7.91	59.1%	0.48	0.58	10.5%
Ours	-6.04	-7.09	-8.43	71.0%	0.43	0.60	24.5%

Security

If you discover a potential security issue in this project, or think you may have discovered a security issue, we ask that you notify Bytedance Security via our security center or vulnerability reporting email.

Please do not create a public GitHub issue.

License

This project is licensed under the Creative Commons Attribution-NonCommercial 4.0 International Public License.