ccdc_roche_scoring

For template based docking and RF-PLP scoring function please cite:

Tosstorff, A. et al. A high quality, industrial data set for binding affinity prediction: performance comparison in different early drug discovery scenarios. J. Comput. Aid. Mol. Des. 36, 753–765 (2022).

For redocking and molecular recognition of docking vs experimental data please cite:

Tosstorff, A., Kuhn, B. Comparison of molecular recognition in docking versus experimental CSD and PDB data.

---

Installation

conda env create -f environment.yaml -p /path/to/env/ccdc_roche_scoring
conda activate ccdc_roche_scoring
poetry install

Pre-commit hook

In order to get in line with current commit conventions of the team, developers can set up a pre-commit hook that is triggered before every commit.

To first install it:

# on the conda env
$> pip install pre-commit
$> pre-commit install
...
$> git commit -am "message" # the hooks will trigger

---

Template based docking in Python

# Folder Structure should look like this:
docking_dir
 |-apo_proteins: contains protein and water files: strucid_dry.mol2, strucid_dry.pdb, strucid_water_0.mol2,...
 |-tmp_aligned_3d_sdf_sanitized
 | # Contains the ligand templates: native_ligands.sdf
 | # Manual curation of the templates is highly recommended to ensure proper tautomers, avoid low quality structures
 |
 |-docking: contains docking_job_0,...

# Dock ligands.sdf, with default target settings. Treat Met713 and Met712 flexibly
cd docking_dir/docking
docking.py --input_ligands ligands.sdf -t default -fr=Met713 Met712

# Concatenate results
join_docked_rf_counts.py -t [target]

---

Template based docking from MOE

One can either dock into MOE, one of the prepared projects or into an MDB of binding sites. The binding sites MDB should have the columns

• 'moe': receptor-ligand complex
• 'ligand': 2D ligand
• 'code': STRUCID

If one of the defined projects is selected, the return results will include the RF-PLP score.

---

Generating project specific parameters for RF-PLP:

• First, run template docking from MOE checking the verbose box. Docked ligands have to contain attribute with pIC50.
• In the docking directory run stat_potential.py

The script will check for the presence of series definitions. If '../../series_assignment.csv' exists, series will be assigned and parameters will be fitted per series. Alternatively, a CSV with columns SRN and split can be passed. The model will then be fitted to the train and validation compounds, not the test compounds.

# Change to your docking directory
cd /path/to/docking_dir
mkdir first iteration

# Collect results from template docking
join_docked_rf_counts.py

# fit RF-PLP
cd first_iteration
stat_potential.py --target [target_name] --task [task_name_pic50]

---

Score with RF-PLP:

• Series definitions should be stored in ccdc_roche_scoring/series_definitions/[target].json
• Model parameters should be stored in ccdc_roche_scoring/scoring_parameters/[target]

ccdc_roche_scoring/stat_potential_inference.py -l ligand.sdf -p protein.pdb -t target_name

---

Applications for MOE

• Start a webserver on the sHPCS:

moeweb -load soap_scoring_client.svl -load soap_template_docking_client.svl

• Adapt the SVL scripts to point to the webserver:
  const SERVER_URL = 'server address';

---

Scoring in MOE

Score an MDB of ligand poses for the active protein.

---