search

durrantlab / autogrow4

AutoGrow4 is an open-source program for semi-automated computer-aided drug discovery. It uses a genetic algorithm to evolve predicted ligands on demand and so is not limited to a virtual library of pre-enumerated compounds. AutoGrow4 is a useful tool for generating entirely novel drug-like molecules and for optimizing preexisting ligands.
http://durrantlab.com/autogrow4/
Apache License 2.0
34 stars 10 forks source link

readme

AutoGrow4

AutoGrow4 is an evolutionary algorithm that optimizes candidate ligands for predicted binding affinity and other drug-like properties. Though no substitute for the medicinal chemist, AutoGrow4 attempts to introduce some chemical intuition into the automated optimization process.

AutoGrow4 takes advantage of recent advancements in multiprocessing, cheminformatics, and docking. It handles all ligand manipulations using SMILES strings via the RDKit API. Docking programs require 3D small-molecule models, so AutoGrow4 use Gypsum-DL to convert the SMILES to 3D.

AutoGrow4 was designed with modularity in mind. New options allow users to control drug-likeness, access new reaction libraries, use additional docking programs, rank AutoGrow4-generated compounds using alternate schemes, and advance candidate compounds via several selection algorithms. The codebase has been redesigned so users can easily add their own custom options as well. Please see the tutorial (/autogrow4/tutorial/TUTORIAL.md), which describes how to expand these various options.

The default fragment libraries included with AutoGrow4 were derived from a subset of the ZINC database (https://zinc.docking.org/). We thank ZINC for allowing us to distribute these fragment libraries to AutoGrow4 users.

Getting Started with AutoGrow4

Use the run_autogrow.py script to run AutoGrow4. The script supports both command-line parameters and a JSON parameter file. A detailed tutorial describing AutoGrow4's dependencies and general use is located at /autogrow4/tutorial/TUTORIAL.md. Additional descriptions of all AutoGrow4 parameters can be obtained by running: python run_autogrow.py -h.

We strongly recommend running AutoGrow4 via Docker using /autogrow4/docker/autogrow_in_docker.py. See the tutorial at /autogrow4/tutorial/TUTORIAL.md for more details.

Dependencies Notes

AutoGrow4 version 4.0.1 has been tested with the following dependencies:

>>> rdkit.__version__
'2020.03.1'
>>> numpy.__version__
'1.18.1'
>>> scipy.__version__
'1.4.1'
>>> matplotlib.__version__
'3.2.1'
>>> func_timeout.__version__
'4.3.5'

If you are unable to run AutoGrow4, please try running AutoGrow4 in a python environment with these specific dependencies. Alternatively, the Docker version of AutoGrow4 automatically installs dependencies that are verified to work with AutoGrow4. If you discover that AutoGrow4 is no longer compatible with current library releases, please contact us, and we will attempt to correct the code.

Developer Note

Dimorphite is Version 1.2.3 with the citation print statements silenced. Please remember to cite Dimorphite-DL: Ropp PJ, Kaminsky JC, Yablonski S, Durrant JD (2019) Dimorphite-DL: An open-source program for enumerating the ionization states of drug-like small molecules. J Cheminform 11:14. doi:10.1186/s13321-019-0336-9.

As of May 2020, AutoGrow4 works in the provided Docker container (with the specific, hard-coded versions of the dependencies). These dependencies are known to be compatible with AutoGrow4. When developing future versions of AutoGrow4, be sure to test the current versions of these dependencies and to update $PATH/docker/Dockerfile appropriately.