This open-source deep docking pipeline is built upon an original Deep Docking pipeline created by Gentile et al. (original paper).
To see detailed instructions on how to run the protocol, please refer to the Open-Source Deep Docking.ipynb notebook provided in this repository. These instructions are complementary to the instructions in the original protocol. The optional downstream analysis is described in Clustering_and_downstream_analysis.ipynb. The instructions are adjusted to use on CSD3 cluster provided by the University of Cambridge, however, can be easily changed to work on any platform.
The pipeline was run on a high-performance computing server of the University of Cambridge (CSD3), however, it can be easily run on any server with sufficient resources. This server provides
To use this protocol, please download the repository or clone it, for example using
git clone https://github.com/MichaelaBrezinova/open_source_deep_docking_protocol.git
This should take under 1 minute.
The protocol was originally run on server nodes using RHEL 7/8 operating system and SLURM. However, it should not be a problem to adapt it to run on different operating systems and job schedulers.
DD_protocol.yml (main), DD_protocol_tensor.yml (for model training, with different version of tensorflow required by A100 GPU units, can be combined with DD_protocol environment or omitted if not needed) and DD_protocol_py27.yml (for use of ChemFp 1.x) contain exported environments that were used by us during the run of the pipeline and are referenced in aforementioned jupyter notebooks.
The main dependencies are similar to those of the original protocol,
- rdkit
- tensorflow >= 1.14.0 (1.15 GPU version recommended. If you are using cuda11, please use nvidia-tensorflow)
- pandas
- numpy
- keras
- matplotlib
- scikit-learn
The protocol uses AutoDock Vina/Vina-GPU and Open Babel.
For downstream analysis, chemfp 1.x, gaucamol are required as well as installation and use of DeePred-BBB and FRED. Chemfp 1.x supports only Python 2.7, hence to use it, an environment with Python 2.7 has to be used (e.g. DD_protocol_py27.yml). These are, however, optional if user wants to omit or adjust these steps.
The installation times for all these tools are standard ( <5 mins). Some packages might take longer if compiled from source (e.g. Open Babel) rather than binary, however, not too long (<30 mins).
DD-prepared version (provided with the original protocol) of the ZINC20 library (as available in March 2021) is available at https://files.docking.org/zinc20-ML/. This data can be further filtered and subsampled per user's needs. An example subset of the library with 10 million compounds is available here https://doi.org/10.6084/m9.figshare.25934092, however, it is advised to use a (much) larger set for the protocol (or if not much larger then well curated) to get good results.
An example target receptor receptor.pdbqt with configuration file conf.txt required by Vina and parameter file logs.txt required by the pipeline are available in the results directory. Parameter file needs to be adjusted based on the user's paths. Open-Source Deep Docking.ipynb contains workflow using this example. An example output after one iteration (iteration 5 from our run) for DD-prepared library (filtered by molecular weight <=360 and prepared by our scripts) is available at https://doi.org/10.6084/m9.figshare.25928428. Individual SDF/PDBQT files for docking of the downloaded compounds are not present, however, can be easily generated from bulk downloaded SDFs that are included. Example provided with the original protocol can also be relevant.
The duration of a full run (in our case it was 5 iterations, possible to change) depends on available hardware, cluster waiting times and number of molecules docked in each iteration. However, on average it should take around 1-2 weeks (in human terms, not computer time).
To cite the original papers by Gentile et al., please use:
@article{gentile2020deep,
title={Deep docking: a deep learning platform for augmentation of structure based drug discovery},
author={Gentile, Francesco and Agrawal, Vibudh and Hsing, Michael and Ton, Anh-Tien and Ban, Fuqiang and Norinder, Ulf and Gleave, Martin E and Cherkasov, Artem},
journal={ACS central science},
volume={6},
number={6},
pages={939--949},
year={2020},
publisher={ACS Publications}
}
@article{gentile2022artificial,
title={Artificial intelligence--enabled virtual screening of ultra-large chemical libraries with deep docking},
author={Gentile, Francesco and Yaacoub, Jean Charle and Gleave, James and Fernandez, Michael and Ton, Anh-Tien and Ban, Fuqiang and Stern, Abraham and Cherkasov, Artem},
journal={Nature Protocols},
volume={17},
number={3},
pages={672--697},
year={2022},
publisher={Nature Publishing Group UK London}
}