TAGMol: Target-Aware Gradient-guided Molecule Generation

[Paper]

---

Environment Setup

The code has been tested in the following environment:

conda create -n tagmol python=3.8.17
conda activate tagmol
conda install pytorch=1.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia
conda install pyg=2.2.0 -c pyg
conda install rdkit=2022.03.2 openbabel=3.1.1 tensorboard=2.13.0 pyyaml=6.0 easydict=1.9 python-lmdb=1.4.1 -c conda-forge

# 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

IMPORTANT NOTE: You might have to do the following to append the path of the root working directory.

export PYTHONPATH=".":$PYTHONPATH

---

Data and Checkpoints

The resources can be found here. The data are inside data directory, the backbone model is inside pretrained_models and the guide checkpoints are inside logs.

Training

Training Diffusion model from scratch

python scripts/train_diffusion.py configs/training.yml

Training Guide model from scratch

BA

python scripts/train_dock_guide.py configs/training_dock_guide.yml

QED

python scripts/train_dock_guide.py configs/training_dock_guide_qed.yml

SA

python scripts/train_dock_guide.py configs/training_dock_guide_sa.yml

NOTE: The outputs are saved in logs/ by default.

---

Sampling

Sampling for pockets in the testset

BackBone

python scripts/sample_diffusion.py configs/sampling.yml --data_id {i} # Replace {i} with the index of the data. i should be between 0 and 99 for the testset.

We have a bash file that can run the inference for the entire test set in a loop.

bash scripts/batch_sample_diffusion.sh configs/sampling.yml backbone

The output will be stored in experiments/backbone. The following variables: BATCH_SIZE, NODE_ALL, NODE_THIS and START_IDX, can be modified in the script file, if required.

BackBone + Gradient Guidance

python scripts/sample_multi_guided_diffusion.py [path-to-config.yml] --data_id {i} # Replace {i} with the index of the data. i should be between 0 and 99 for the testset.

To run inference on all 100 targets in the test set:

bash scripts/batch_sample_multi_guided_diffusion.sh [path-to-config.yml] [output-dir-name]

The outputs are stored in experiments_multi/[output-dir-name]when run using the bash file. The config files are available in configs/noise_guide_multi.

• Single-objective guidance
  • BA: sampling_guided_ba_1.yml
  • QED: sampling_guided_qed_1.yml
  • SA: sampling_guided_sa_1.yml
• Dual-objective guidance
  • QED + BA: sampling_guided_qed_0.5_ba_0.5.yml
  • SA + BA: sampling_guided_sa_0.5_ba_0.5.yml
  • QED + SA: sampling_guided_qed_0.5_sa_0.5.yml
• Multi-objective guidance (our main model)
  • QED + SA + BA: sampling_guided_qed_0.33_sa_0.33_ba_0.34.yml

For example, to run the multi-objective setting (i.e., our model):

bash scripts/batch_sample_multi_guided_diffusion.sh configs/noise_guide_multi/sampling_guided_qed_0.33_sa_0.33_ba_0.34.yml qed_0.33_sa_0.33_ba_0.34

---

Evaluation

Evaluating Guide models

python scripts/eval_dock_guide.py --ckpt_path [path-to-checkpoint.pt]

Evaluation from sampling results

python scripts/evaluate_diffusion.py {OUTPUT_DIR} --docking_mode vina_score --protein_root data/test_set

The docking mode can be chosen from {qvina, vina_score, vina_dock, none}

NOTE: It will take some time to prepare pqdqt and pqr files when you run the evaluation code with vina_score/vina_dock docking mode for the first time.

---

Results

Methods	Vina Score (↓)		Vina Min (↓)		Vina Dock (↓)		High Affinity (↑)		QED (↑)		SA (↑)		Diversity (↑)		Hit Rate % (↑)
	Avg.	Med.	Avg.	Med.	Avg.	Med.	Avg.	Med.	Avg.	Med.	Avg.	Med.	Avg.	Med.
Reference	-6.36	-6.46	-6.71	-6.49	-7.45	-7.26	-	-	0.48	0.47	0.73	0.74	-	-	21
liGAN	-	-	-	-	-6.33	-6.20	21.1%	11.1%	0.39	0.39	0.59	0.57	0.66	0.67	13.2
AR	-5.75	-5.64	-6.18	-5.88	-6.75	-6.62	37.9%	31.0%	0.51	0.50	0.63	0.63	0.70	0.70	12.9
Pocket2Mol	-5.14	-4.70	-6.42	-5.82	-7.15	-6.79	48.4%	51.0%	0.56	0.57	0.74	0.75	0.69	0.71	24.3
TargetDiff	-5.47	-6.30	-6.64	-6.83	-7.80	-7.91	58.1%	59.1%	0.48	0.48	0.58	0.58	0.72	0.71	20.5
DecompDiff	-4.85	-6.03	-6.76	-7.09	-8.48	-8.50	64.8%	78.6%	0.44	0.41	0.59	0.59	0.63	0.62	24.9
TAGMol	-7.02	-7.77	-7.95	-8.07	-8.59	-8.69	69.8%	76.4%	0.55	0.56	0.56	0.56	0.69	0.70	27.7

Due to space constraints, we only share the eval_results folder generated from the evaluation script. It can be found in the same link as other resources, inside results directory.

---

Citation

@article{dorna2024tagmol,
  title={TAGMol: Target-Aware Gradient-guided Molecule Generation},
  author={Vineeth Dorna and D. Subhalingam and Keshav Kolluru and Shreshth Tuli and Mrityunjay Singh and Saurabh Singal and N. M. Anoop Krishnan and Sayan Ranu},
  journal={arXiv preprint arXiv:2406.01650},
  year={2024}
}

---

Acknowledgements

This codebase was build on top of TargetDiff