SCORCH

---

SCORCH (Scoring COnsensus for RMSD-based Classification of Hits) is a fast scoring function based on a consensus of machine learning models. Scoring functions are used to evaluate poses of molecules obtained from molecular docking. SCORCH scores range from 0 to 1, with higher values indicating a higher probability of the molecule binding tightly to the receptor.

SCORCH uses .pdbqt files as input for the scoring, which is the format used by Autodock, Vina, and GWOVina docking software, among others. Additionally, this release contains an integrated pipeline to dock and score molecules in SMILES format using GWOVina.

SCORCH uses a variety of descriptors to characterize a docked pose, including Binana 1.3 and ECIFs. The contributing models were trained on multiple docked poses for each ligand, labelled based on their RMSD to crystal structures. Training examples of non-binders were generated using DeepCoy. SCORCH has used over 54,000 poses in its training. As a result, SCORCH avoids biases and provides improved accuracy to identify true binder molecules in virtual screening. Read more in our publication.

---

Installation

Linux

Installation on Linux is achieved via conda. The supplied setup bash script installs SCORCH and dependencies in a conda environment, MGLTools 1.5.6 and GWOVina 1.0. If you do not have conda installed, the setup bash script installs it for you silently in the SCORCH directory.

To install SCORCH on Linux:

# clone the GitHub repository
git clone https://github.com/SMVDGroup/SCORCH.git

# ensure setup.sh is executable
cd SCORCH
sudo chmod +x setup.sh

# execute the setup script
./setup.sh

Mac OS

Due to MGLTools-1.5.6 conflicts with newer Mac OS versions and M1 chips, SCORCH and SCORCH's SMILES docking and scoring pipeline can only be run on Mac OS systems via Docker which can be installed here. Docker can also be used to run SCORCH on Linux systems in case of problems with the installation script.

To set up SCORCH once Docker is installed:

# download the scorch docker image
docker pull scorchml/scorch:v1.0

# run the scorch docker image
docker run -i -t scorchml/scorch:v1.0 /bin/bash

# once inside the docker image
cd home/SCORCH

Receptor & Ligand Preparation

The scoring function accepts .pdbqt receptor files and SMILES or .pdbqt ligand files as inputs. Any .pdb receptor files should be prepared with the supplied MGLTools 1.5.6 using prepare_receptor4.py Python script as follows:

# preparing a receptor
./utils/MGLTools-1.5.6/bin/pythonsh \
./utils/MGLTools-1.5.6/MGLToolsPckgs/AutoDockTools/Utilities24/prepare_receptor4.py \
-r examples/predocked_1a0q/1a0q_receptor.pdb \
-A hydrogens \
-o examples/predocked_1a0q/1a0q_receptor_out.pdbqt \
-U nphs

SMILES ligands need no preprocessing. For pre-docked .pdbqt ligands, we recommend only scoring docking results in .pdbqt format (ideally from AutoDock or GWOVina). Scoring results from other docking software might be possible if converted to .pdbqt.

Usage

To use the scoring function on both Linux and Mac OS, the conda environment needs to be activated first:

conda activate scorch

The scoring function is supplied as the Python script scorch.py. Its main arguments are:

Argument	Value	Importance
-r, --receptor	Filepath to receptor file (pdbqt)	Essential
-l, --ligand	Filepath to ligand(s) (pdbqt or SMILES)	Essential
-rl, --ref_lig	Filepath to example ligand in receptor binding site (mol, mol2, sdf, pdb or pdbqt)	Essential for SMILES ligands (unless --center and --range supplied)
-c, --center	'[x, y, z]' coordinates of the center of the binding site for docking	Essential for SMILES ligands (unless --ref_lig supplied)
-ra, --range	'[x, y, z]' axis lengths to define a box around --center coordinates for docking	Essential for SMILES ligands (unless --ref_lig supplied)
-o, --out	Filepath for output csv (If not supplied, scores are written to stdout)	Optional (Default stdout)
-p, --return_pose_scores	If supplied, scoring values for individual poses in each ligand file are returned	Optional (Default False)
-t, --threads	Number of threads to use	Optional (Default 1)
-v, --verbose	If supplied, progress bars and indicators are displayed while scoring	Optional (Default False)

For further details on arguments, run python scorch.py --h.

Docking and Scoring SMILES Ligands Against a Receptor

SCORCH includes a full pipeline to convert SMILES ligands to .pdbqt files using MGLTools 1.5.6, dock them using GWOVina, and score them with SCORCH:

python scorch.py \
--receptor examples/smiles_REPTIN/reptin_receptor.pdbqt \
--ligand examples/smiles_REPTIN/reptin_smiles.smi  \
--ref_lig examples/smiles_REPTIN/reptin_ref_lig.pdbqt \
--out scoring_results.csv

The binding site for docking can be defined with a reference ligand as above with --ref_lig, or by supplying --center and --range values in the same way as for molecular docking software:

python scorch.py \
--receptor examples/smiles_REPTIN/reptin_receptor.pdbqt \
--ligand examples/smiles_REPTIN/reptin_smiles.smi  \
--center '[23.981,-42.667,67.156]' \
--range '[16.5,14.25,16.5]' \
--out scoring_results.csv

The --ligand input should be supplied as .smi or .txt file, with one SMILES ligand and an optional identifier per line, as in this example:

CCC(CC)O[C@@H]1C=C(C[C@H]([C@H]1NC(=O)C)O)C(=O)OCC 49817880
CCC(CC)O[C@@H]1C=C(C[C@@H]([C@H]1NC(=O)C)[NH3+])C(=O)OCC 24848267
CCC(CC)O[C@@H]1C=C(C[C@@H]([C@H]1NC(=O)C)N)C(=O)NOC 118722031

Docking settings can be changed by editing the utils/params/dock_settings.json file in the scoring function folder. See the function help for additional details.

Scoring Already Docked Ligands Against a Receptor

For scoring a single ligand - examples/predocked_1a0q/ligands/1a0q_docked_ligand.pdbqt - against a single receptor - examples/predocked_1a0q/1a0q_receptor.pdbqt:

python scorch.py \
--receptor examples/predocked_1a0q/1a0q_receptor.pdbqt \
--ligand examples/predocked_1a0q/ligands/1a0q_docked_ligand.pdbqt

For scoring all .pdbqt ligands in the directory - examples/predocked_1a0q/ligands/ - against a single receptor - examples/predocked_1a0q/1a0q_receptor.pdbqt - just supply the directory path to the --ligand argument:

python scorch.py \
--receptor examples/predocked_1a0q/1a0q_receptor.pdbqt \
--ligand examples/predocked_1a0q/ligands/ \
# this writes the output scores to a file
--out scoring_results.csv \
# this parallelises the scoring over 6 threads
--threads 6 \
# this displays scoring progress
--verbose

Importing SCORCH as a Python module

The main function from scorch.py can be imported and used in other Python scripts. It takes a dictionary of parameters as inputs and returns a pandas dataframe of model scores identical to the normal scoring function output:

from scorch import scoring, parse_module_args

input_parameters = {'ligand': 'examples/predocked_1a0q/ligands/',
                  'receptor': 'examples/predocked_1a0q/1a0q_receptor.pdbqt',
                  'threads': 4,
                  'verbose': False}

parsed_parameters = parse_module_args(input_parameters)

output = scoring(parsed_parameters)

print(output)

Output

Scores are output in .csv format. For example, scoring a single ligand .pdbqt file with 10 docked poses on a single receptor, using the --return_pose_scores flag, yielded the following output. Note that the output of SCORCH includes a measure of the prediction certainty.

Receptor	Ligand	SCORCH_pose_score	SCORCH_certainty	Ligand_ID	SCORCH_score	best_pose
receptor.pdbqt	ligand_pose_1	0.83521	0.75148	ligand	0.86669	0
receptor.pdbqt	ligand_pose_2	0.83782	0.75926	ligand	0.86669	0
receptor.pdbqt	ligand_pose_3	0.84241	0.74976	ligand	0.86669	0
receptor.pdbqt	ligand_pose_4	0.77493	0.65857	ligand	0.86669	0
receptor.pdbqt	ligand_pose_5	0.72339	0.697	ligand	0.86669	0
receptor.pdbqt	ligand_pose_6	0.86471	0.78335	ligand	0.86669	0
receptor.pdbqt	ligand_pose_7	0.81688	0.69923	ligand	0.86669	0
receptor.pdbqt	ligand_pose_8	0.86669	0.78808	ligand	0.86669	1
receptor.pdbqt	ligand_pose_9	0.65023	0.74499	ligand	0.86669	0
receptor.pdbqt	ligand_pose_10	0.07948	0.86123	ligand	0.86669	0