This protocol is used for performing structural modification of ligands with Delete method.
Download the Delete and follow the guidelines to build the surf_maker environment and delete environment.
Scenario: if you have a target on which to design, and you have a prepared fragment. You can generate a series of possible modifications using Delete. The used checkpoint can be downloaded here.
python gen_all_epoch_to_1_target.pyThen you would get a series of modified molecules.
For a number of reasons, some proteins may not have the experimental conformations, such as parp7. To use Delete for this case, you need to predict its conformation with AlphaFold2 or homology modeling.
Once you get the AlphaFold-predicted conformations, they often contain some unphysical local structures, like twisted rings or something. The geometry optimization is suggested. I have prepared an OpenMM-based code for doing this automatically.
The first thing is to clean the protein file, like replacing nonstandard residues and repairing atoms and residues.
pdbfixer protein.pdb --replace-nonstandard --add-atoms=all --add-residuesThen, perform local geometry optimization.
python ./protein-geom/protein_geom_opt.py --input_pdb xxx --out_name xxx(Optional) You can run the protein conformation simulation with explicit solvent.
python ./protein-geom/protein_geom_opt.py --input_pdb xxx --out_name xxx
(Optional) Sometimes, you may want to perform the protein-ligand simulation to explore more possibilities, like optimizing molecules from different conformational states.
To be continued. You can use a series of docking software, where the scripts are stored in the ./docking
OpenFreeEnergy/Lomap: Alchemical mutation scoring map (github.com)
One way to visualize the generated molecules is by mapping them with the t-map, please refer to the following Git for implementation. The different color could be attributed to different properties.
https://github.com/HaotianZhangAI4Science/AI-Physics-DrugDiscovery/tree/main/chemical_space/tmap
You can use a similarity matrix to cluster molecules, and select several ones in each cluster. I have provided an example in the following repository.
./clustering/molecules_cluster.ipynb
You can use the ./clustering/sort_by_docking.py or ./clustering/sort_by_similarity.ipynb to sort the molecules into PDF format for experts' review. The generated example are outlined as follows:
ChemBio https://chembioserver.vi-seem.eu/Dendrogram.php
Scaffold Hunter. https://scaffoldhunter.sourceforge.net/
chemaxon https://chemaxon.com/discovery-tools
chemmine http://chemmine.ucr.edu/about/
ChemBio is my first choice since it is quite easy to use.
Synthesizing designed molecules is not available for some researchers. Therefore, we provide the script to select the molecule most similar to the generated one. This protocol indeed has some advantages, including securing the drug-like properties of ligands, but it sacrifices the chemical space exploration abilities, sometimes leading to the missing of hit compounds.
There are plenty of choices of ligand libraries, such as CHEMBL, ZINC, ChemDiv, and Specs. You should download them from the original resources in the ./library.
ChemDiv: ChemDiv contains several .sdf files, including DC01_400000.sdf, DC02_400000.sdf, DC03_332268.sdf, IC_136982.sdf, and NC_292464.sdf. You can use obabel to merge them together.
obabel DC01_400000.sdf, DC02_400000.sdf, DC03_332268.sdf, IC_136982.sdf, NC_292464.sdf -O chemdiv.sdfSpecs: Specs only has one file, whereas mine is named Specs_ExAcD_Aug_2020.sdf.
ZINC: Zinc needs you to download the files, where the command is obtained from the ZINC Website. I have prepared a download.sh makingscript in the ./library/Zinc_download.ipynb.
Our script employs commercial software, OpenEye, for fast-searching compound libraries.
# Make fingerprints of libraries. Here we show the chemdiv example. Try other libraries based on your own needs.
python ./openeye/makefastfp.py -in ./library/chemdiv.sdf -fpdb ./library/chemdiv.fpbin
# Once you obtain the fingerprint database, you can search the query mol using the following script.
python ./openeye/searchfastfp.py -fpdb ./library/chemdiv.fpbin -molfname ./library/chemdiv.sdf -query ./library/query.sdf -out hits.sdf -memorytype in-memorypython ./openeye/search_substructure.py --query_sdf query.sdf --library ./library/chemdiv.sdf --save_file ./hit.sdf