Skip to content

Computationally Accelerating Protein-Ligand Matching: A Case Study on Leishmaniasis

Notifications You must be signed in to change notification settings

wl-research/docking-wl

 
 

Repository files navigation

Accepted Poster at the MLPC Workshop of ICLR 2021 : poster

We present in this work the intersection of Deep Learning-based and empirical Molecular docking tools for the ligand-target interaction. The first part of the work relies on Deep Learning, in this case, we used DeepPurpose , the second part on Molecular docking tool Autodock Vina. The molecular docking involves two main internal steps : the extraction of the most promising binding sites from the protein structures, and the computation of the energy of the pose made of the ligand located in the most promising protein's binding sites. After computing the pose's energies with autodock vina, we optimzed them by setting the ligand in their corresponding binding sites using chimera

Requirements

  • DeepPurpose : the instructions are described in this repository
  • Autodock Vina & Openbabel
sudo apt update
sudo apt install autodock-vina
sudo apt-get install -y openbabel

Deep Learning Ligand-Target Affinity Computation

The ligand-target affinity computation consisted in the following steps:

  • ligand & target conversion using openbabel
    • Ligand : From pdb to smiles
    • Target : From pdb to Amino Acid Sequence
  • Binding-score computation : the Deepurpose pretrained models were used to figured out the affinity from the data (ligand & target) previously well converted. This code was used to carry out this operation.

Molecular Docking

In this part, we performed the following operations to realize the protein-ligand docking.

  • Ligand-Target preparation: Autodock vina required ligands and target both converted into pdbqt format. To realize it, we used the Vina's built-in scripts to prepare both ligands and proteins.
  • Once the ligand and receptor preparation completed, we moved first, on searching the binding pockets of the receptor then on computing the energy of the ligand-receptor pairs with ligand located in each of the most promising binding sites. This was done using CB-Dock built on Autodock vina. The main added feature of CB-Dock is the computation of the receptor binding pocket, the remain work related to the docking entirely relies on Autodock (vina).
  • In the last step, we built the optimized pose (ligand-receptor) using chimera which readily output the optimized pose into pdb format for submission at the indaba-grand-challenge-curing-leishmaniasis challenge

Strategy for pose selection

Run the docking with the pair-wise combinations of all ligands and receptors will be computationally expensive, to overpass this we came out with a strategy to downside the list of pairs. The main points are described in the following pseudo-code

# X in [100(100)1400]
for target in targets:
   ligands = select_top_x(target) # select the top x drugs according to the binding scores
   for ligand in ligandss:
       run docking(target, ligand)
   ligands = sorted(ligandss) # according to the energy from docking
   best_target = search_best_target(targets, ligands) # search the pair for which ligands most appear as the top ligand
   pairs = make_pair(best_target, ligands)
   submit(pairs) ## submit the pairs to indaba-grand-challenge

Docking Protocol

Foremost, the protocol for the ligand-protein affinity computation is well describe in the DeepPurpose's repository. We hereby describe only the guidelines for the docking protocol with the environment settled.

Assumptions

  • The folder protein-pdbqt contains the pdbqt version of the proteins we used in this experiment.
  • The folder ligand-pqbqt contains the pdbqt version of the small molecules we used.
  • The folder configs contain the best three binding sites for each protein. The following command performs the molecular docking for each pair in all the files contained in the folder binding_score
python3 docking.py --output 'outputs/' --binding_score 'binding_scores/' --config 'configs/' --log 'logs/' --ligand 'ligand-pdbqt/' --protein 'protein-pdbqt/'

or

bash docking.sh

About

Computationally Accelerating Protein-Ligand Matching: A Case Study on Leishmaniasis

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 97.8%
  • Python 2.1%
  • Shell 0.1%