EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction

Paper on arXiv

EquiBind, is a SE(3)-equivariant geometric deep learning model performing direct-shot prediction of both i) the receptor binding location (blind docking) and ii) the ligand’s bound pose and orientation. EquiBind achieves significant speed-ups and better quality compared to traditional and recent baselines. If you have questions, don't hesitate to open an issue or ask me via [email protected] or social media or Octavian Ganea via [email protected]. We are happy to hear from you!

Dataset

Our preprocessed data (see dataset section in the paper Appendix) is available from zenodo.
The files in data contain the names for the time-based data split.

If you want to train one of our models with the data then:

1. download it from zenodo
2. unzip the directory and place it into data such that you have the path data/PDBBind

Use provided model weights to predict binding structure of your own protein-ligand pairs:

Step 1: What you need as input

Ligand files of the formats .mol2 or .sdf or .pdbqt or .pdb whose names contain the string ligand (your ligand files should contain all hydrogens).
Receptor files of the format .pdb whose names contain the string protein. We ran reduce on our training proteins. Maybe you also want to run it on your protein.
For each complex you want to predict you need a directory containing the ligand and receptor file. Like this:

my_data_folder
└───name1
    │   name1_protein.pdb
    │   name1_ligand.sdf
└───name2
    │   name2_protein.pdb
    │   name2_ligand.mol2
...

Step 2: Setup Environment

We will set up the environment using Anaconda. Clone the current repo

git clone https://github.com/HannesStark/EquiBind

Create a new environment with all required packages using environment.yml. If you have a CUDA GPU run:

conda env create -f environment.yml

If you instead only have a CPU run:

conda env create -f environment_cpuonly.yml

Activate the environment

conda activate equibind

Here are the requirements themselves for the case with a CUDA GPU if you want to install them manually instead of using the environment.yml:

python=3.7
pytorch 1.10
torchvision
cudatoolkit=10.2
torchaudio
dgl-cuda10.2
rdkit
openbabel
biopython
rdkit
biopandas
pot
dgllife
joblib
pyaml
icecream
matplotlib
tensorboard

If Docker is more your speed for assembling the operating environment, a docker image with docker compose functionality for local bind mounting is included. To set up, assuming you have docker desktop set up, navigate to the main directory where the dockerfiles are located and run:

docker-compose run --rm app

Step 3: Predict Binding Structures!

In the config file configs_clean/inference.yml set the path to your input data folder inference_path: path_to/my_data_folder.
Then run:

python inference.py --config=configs_clean/inference.yml

Done! 🎉
Your results are saved as .sdf files in the directory specified in the config file under output_directory: 'data/results/output' and as tensors at runs/flexible_self_docking/predictions_RDKitFalse.pt!

Inference for multiple ligands in the same .sdf file and a single receptor

python multiligand_infernce.py -o path/to/output_directory -r path/to/receptor.pdb -l path/to/ligands.sdf

This runs EquiBind on every ligand in ligands.sdf against the protein in receptor.pdb. The outputs are 3 files in output_directory with the following names and contents:

failed.txt - contains the index (in the file ligands.sdf) and name of every molecule for which inference failed in a way that was caught and handled.
success.txt - contains the index (in the file ligands.sdf) and name of every molecule for which inference succeeded.
output.sdf - contains the conformers produced by EquiBind in .sdf format.

Reproducing paper numbers

Download the data and place it as described in the "Dataset" section above.

Using the provided model weights

To predict binding structures using the provided model weights run:

python inference.py --config=configs_clean/inference_file_for_reproduce.yml

This will give you the results of EquiBind-U and then those of EquiBind after running the fast ligand point cloud fitting corrections.
The numbers are a bit better than what is reported in the paper. We will put the improved numbers into the next update of the paper.

Training a model yourself and using those weights

To train the model yourself, run:

python train.py --config=configs_clean/RDKitCoords_flexible_self_docking.yml

The model weights are saved in the runs directory.
You can also start a tensorboard server tensorboard --logdir=runs and watch the model train.
To evaluate the model on the test set, change the run_dirs: entry of the config file inference_file_for_reproduce.yml to point to the directory produced in runs. Then you can runpython inference.py --config=configs_clean/inference_file_for_reproduce.yml as above!

Reference

📃 Paper on arXiv

@misc{stark2022equibind,
      title={EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction}, 
      author={Hannes Stärk and Octavian-Eugen Ganea and Lagnajit Pattanaik and Regina Barzilay and Tommi Jaakkola},
      year={2022}
}