POEM: Pocket-Oriented Elaboration of molecules

Description

This project aims at using information from ligand pieces bound to protein subpockets to (semi-) automatically build new molecules tailored to a particular target pocket on the basis of the subpockets/target pocket estimated similarities.
This workflow was tested to design new sub-micromolar hit candidates for CDK8 inhibition:

Eguida M., Schmitt-Valencia C., Hibert M., Villa, P. and Rognan, D. Target-Focused Library Design by Pocket-Applied Computer Vision and Fragment Deep Generative Linking. J. Med. Chem. 2022, 65, 13771–13783. https://doi.org/10.1021/acs.jmedchem.2c00931

Please note that the publication refers to release v1.0.0.

Content

• envs/ ---> conda environments
• cdk8_structures/ ---> target structures
• scripts/ ---> scripts for library generation
• aligned_fragments.tgz downloadable at 10.5281/zenodo.7023191 ---> output data after steps 1-3, input for step 4+
• output_files.tgz downloadable at 10.5281/zenodo.7023191 ---> data obtained at each step, and depedencies

Requirements

• Conda: https://docs.conda.io/projects/conda/en/latest/user-guide/install/linux.html environements with python 3.6+.
• IChem: http://bioinfo-pharma.u-strasbg.fr/labwebsite/download.html
• ProCare: https://github.com/kimeguida/ProCare
• DeLinker: https://github.com/oxpig/DeLinker
• RDKit: http://www.rdkit.org, https://github.com/rdkit/rdkit

Part 1/ Subpocket screening and fragments preparation

1. Alignment of fragments'supocket clouds to target pocket cloud with ProCare

sc-PDB subpockets and fragments

• input structures from the sc-PDB database
  
• fragmentation, cavity clouds and interactions computed with IChem
   for more information on fragmentation, sc-PDBFrag
  

CDK8 pocket
<this_repo>/cdk8_structures/5hbh_cavityALL_p0-p1-p6.mol2

ProCare: https://github.com/kimeguida/ProCare (state of our conda env: envs/procare.yml)

source <path_to_your_conda>
conda activate procare
cd aligned_fragments/
python ../scripts/procare_launcher.py -s <subpocket> -t ../cdk8_structures/5hbh_cavityALL_p0-p1-p6.mol2 --transform --ligandtransform <fragment>

Outputs:

• aligned subpockets, fragments and procare_scores.tsv available on zenodo
• subpocket: cfh_xx_fragN_cavity4.mol2
• fragment: cfh_xx_fragN.mol2

2. Convert aligned fragments from mol2 to sdf

We used OpenEye python toolkits (state of our conda env: envs/oepython.yml):
conda deactivate
conda activate oepython
../scripts/convert.py <fragment>.mol2 <fragment>.sdf

Outputs:

• sdf available on zenodo
• fragment: cfh_xx_fragN.sdf

3. Compute IChem interactions:

<path_to_your_ichem>/IChem ../cdk8_structures/5hbh_protein.mol2 <fragment>.mol2 > <fragment>.ifp

Outputs:

• interactions available on zenodo
• interactions file: cfh_xx_fragN.ifp

Part 2/ Selection and annotation of relevant fragments

To reproduce this step, the required data out of steps 1-to-3 were made availaible at https://zenodo.org/record/7023191 as aligned_fragments.tgz and ouputs obtained output_files.tgz

4. Select top-scored subpockets and annotate fragments according to six predefined CDK8 areas

tar -xzf aligned_fragments.tgz
cd aligned_fragments/

current directory: aligned_fragments containing data from steps 1-3

assignment of CDK8 areas
conda deactivate
conda activate delinker (state of our conda env: envs/delinker.yml)
python ../scripts/select_fragments_round1.py -f ../output_files/procare_scores.tsv -d . -p ../cdk8_structures/5hbh_protein.mol2 -c ../cdk8_structures/5hbh_cavityALL_p0-p1-p6.mol2

Outputs:

• subpocket_p0_gate.list which corresponds to GA1 in paper
  
• subpocket_p0_hinge.list ---> H
  
• subpocket_p0_solv_1.list ---> SE2
  
• subpocket_p0_solv_2.list ---> SE1
  
• subpocket_p6_alphaC.list ---> AC
  
• subpocket_p6_lys52.list ---> GA2
  available in <this_repo>/output_files/
  

Part 3/ Fragments linking

To reproduce these steps, the required data were made availaible at https://zenodo.org/record/7023191 as aligned_fragments.tgz and ouputs obtained output_files.tgz

5. Enumerate candidates for linking: pairs of fragments and atoms

python ../scripts/linkable_fragments_round1_job.py --hinge subpocket_p0_hinge.list --gate subpocket_p0_gate.list --solv1 subpocket_p0_solv_1.list --solv2 subpocket_p0_solv_2.list --alphac subpocket_p6_alphaC.list --lys52 subpocket_p6_lys52.list

Outputs:

• linkable_fragments_round1_<N>.list with N in {0, 1, 2, 3, 4, 5, 6} available in <this_repo>/output_files/
  

6. Linking with DeLinker

DeLinker: https://github.com/oxpig/DeLinker

Feed DeLinker with connecatble candidates:
python ../scripts/delinker.py -f linkable_fragments_round1_<N>.list -p <your_path_to_DeLinker>/DeLinker/ > /dev/null
This step was distributed on computer clusters.
Output: generation.smi renamed and zipped as generation_<N>.smi.gz with N in {0, 1, 2, 3, 4, 5, 6} available in <this_repo>/output_files/

7. Check generation success

Sometimes, no linker is generated and DeLinker might return truncated attempts.
python ../scripts/get_linker.py --file generation_<N>.smi.gz --fragsdir . --pathdelinker <your_path_to_DeLinker>/DeLinker/

Outputs:

• generation_complete.smi
  
• generation_uncomplete.smi
  

8. Name molecules and filter with openEye Filter

SMILES were assigned IDs to keep track of the molecules infos. Filter will protonate and generate canonical SMILES different from RDKit's.
python ../scripts/index_generated_molecules.py -i generation_complete.smi -o generation_complete_indexed.smi

Output:

• generation_complete_indexed.smi
  

<path_to_openeye>/filter -in generation_complete_indexed.smi -out druglike_molecules.smi -fail druglike_failed_molecules.smi -filter ../output_files/filter_labo_cdk8.txt

Check that other annotations in the file did not affect how Filter processed the SMILES. In our case, we extracted the SMILES and indexes to a separate file molecules.smi.

Output:

• druglike_molecules.smi
  

9. Synthetic accessibility, descriptors, filtering:

SAscore from https://github.com/rdkit/rdkit/blob/master/Contrib/SA_Score/sascorer.py
python ../scripts/get_sascore.py -i druglike_molecules.smi -o druglike_molecules_sascore.tsv
Output:

• druglike_molecules_sascore.tsv
  

RDKit descriptors
python ../scripts/get_druglike_descriptors.py -i druglike_molecules.smi -o druglike_molecules_descriptors.tsv
python ../scripts/get_linker_descriptors.py -i generation_complete_indexed.smi -o generation_linker_descriptors.tsv

Outputs:

• druglike_molecules_descriptors.tsv
• generation_linker_descriptors.tsv

Clean generated linkers, remove too flexible, hydrophobic
python ../scripts/filter_linker.py -i generation_linker_descriptors.tsv -o linker_discarded.tsv

Output:

• linker_discarded.tsv
  

10. Extract round 1 library

python ../scripts/library_round1.py --descriptor druglike_molecules_descriptors.tsv --sascore druglike_molecules_sascore.tsv --discarded linker_discarded.tsv -o libr1.txt

Output:

• libr1.txt
  

11. Grow molecules in round 2 library from a selected hit

Example of hit compound 12
python ../scripts/round2_fuse_mols.py --dl druglike_molecules.smi --gen generation_complete_indexed.smi --origin ../output_files/frag_origin.tsv --discarded linker_discarded.tsv --procare ../output_files/procare_scores.tsv

Outputs:

• hit12_round2_mols.tsv
  
• hit12_round2_sascore_pass.tsv
  

12. Filter and generate round 2 library

RDKit descriptors
python ../scripts/get_round2_descriptors.py -i hit12_round2_mols.tsv -o hit12_round2_mols_descriptors.tsv

Output:

• hit12_round2_mols_descriptors.tsv
  

candidates for synthesis
python ../scripts/library_round2.py -i hit12_round2_mols_descriptors.tsv --sascore hit12_round2_sascore_pass.tsv -o libr2.txt

Output:

• libr2.txt
  

Help and issues

Signal issues

https://github.com/kimeguida/POEM/issues

Support contacts

Merveille Eguida: keguida'[at]'unistra'[dot]'fr
Didier Rognan, PhD: rognan'[at]'unistra'[dot]'fr

Prospective applications

1. Eguida, M.; Schmitt-Valencia, C.; Hibert, M.; Villa, P.; Rognan, D. Target-Focused Library Design by Pocket-Applied Computer Vision and Fragment Deep Generative Linking. J. Med. Chem. 2022, 65, 13771–13783. https://doi.org/10.1021/acs.jmedchem.2c00931

2. CACHE hits prediction Challenge #1 https://cache-challenge.org

Citation

If you use POEM, please cite:

@article{doi:10.1021/acs.jmedchem.2c00931,
abstract = {We here describe a computational approach (POEM: Pocket Oriented Elaboration of Molecules) to drive the generation of target-focused libraries while taking advantage of all publicly available structural information on protein-ligand complexes. A collection of 31 384 PDB-derived images with key shapes and pharmacophoric properties, describing fragment-bound microenvironments, is first aligned to the query target cavity by a computer vision method. The fragments of the most similar PDB subpockets are then directly positioned in the query cavity using the corresponding image transformation matrices. Lastly, suitable connectable atoms of oriented fragment pairs are linked by a deep generative model to yield fully connected molecules. POEM was applied to generate a library of 1.5 million potential cyclin-dependent kinase 8 inhibitors. By synthesizing and testing as few as 43 compounds, a few nanomolar inhibitors were quickly obtained with limited resources in just two iterative cycles.},
author = {Eguida, Merveille; Schmitt-Valencia, Christel; Hibert, Marcel; Villa, Pascal and Rognan, Didier},
title = {{Target-Focused Library Design by Pocket-Applied Computer Vision and Fragment Deep Generative Linking}},
journal = {Journal of Medicinal Chemistry},
volume = {65},
number = {20},
pages = {13771--13783},
year = {2022},
doi = {10.1021/acs.jmedchem.2c00931},
URL = {https://doi.org/10.1021/acs.jmedchem.2c00931},
}

References

• RDKit: Open-source cheminformatics; http://www.rdkit.org, https://github.com/rdkit/rdkit
• Da Silva, F.; Desaphy, J.; Rognan, D. IChem: A Versatile Toolkit for Detecting, Comparing, and Predicting Protein–Ligand Interactions. ChemMedChem 2018, 13, 507–510.
• Desaphy, J.; Bret, G.; Rognan, D.; Kellenberger, E. Sc-PDB: A 3D-Database of Ligandable Binding Sites—10 Years On. Nucleic Acids Res. 2014, 43, D399–D404.
• Eguida, M.; Rognan, D. A Computer Vision Approach to Align and Compare Protein Cavities: Application to Fragment-Based Drug Design. J. Med. Chem. 2020, 63, 7127–7142.
• Imrie, F.; Bradley, A. R.; van der Schaar, M.; Deane, C. M. Deep Generative Models for 3D Linker Design. J. Chem. Inf. Model. 2020, 60, 1983–1995.