This function defined as RFL-Score was trained using the PDBBind 2018 (general) and CSAR (decoys and NRC HiQ). From these complexes we extract molecular descriptors using BiNANA, RDKit 2D/3D, SASA, PaDEL-Descriptor, Vina, totaling 723 descriptors. Then, we selected the most promising attributes using LassoCV and parametrized a Random Forest (RF) algorithm using the GridSearchCV. This RF model was validated using CASF 2013 and 2016.
- Python 3.7+
- MGLTools 1.5.6
- vina4dv
- DSSP 4.0.4
- JRE 11.0.10+
- Pandas 1.1.3+
- Biopython 1.78+
- RDKit 2020.09.1+
- OpenBabel 2.4
- Joblib 0.17.0+
- Scikit-learn 0.23.2+
- ODDT 0.7+
- R 4.0.3
- Random Forest Package 4.6.14+
The current scoring function is executed with Python 3.7. The most simple way to install the SF is by creating an Anaconda Environment. To download the last version of Anaconda Individal Edition visit his official page: https://www.anaconda.com/produacts/individual
A proper environment in Anaconda is useful to install the SF. To create a new environment with Python 3.7 execute the following statement in Anaconda's prompt:
conda create -n 3_7 python=3.7
To activate the new environment execute the following statement:
source activate 3_7
The new environment can be activated automatically adding the previous statement in the .bashrc file.
Pandas It is a package for data analysis and manipulation tool. To install in Conda execute the following statement:
conda install -c anaconda pandas
Biopython It is a set of freely available tools for biological computation. To install in Conda execute the following statement:
conda install -c anaconda biopython
RDKit It is a cheminformatics software to create 2D and 3D molecular features for machine learning applications. To install in Conda execute the following statement:
conda install -c rdkit rdkit
OpenBabel 2.4 It is a chemical software designed to speak the many languages of chemical data. The version 2.4 is mandatory to install the SF. To install in Conda execute the following statement:
conda install -c openbabel openbabel
ODDT (Open Drug Discovery Toolkit) It is a modular and comprehensive toolkit for use in cheminformatics, molecular modeling etc. To install in Conda execute the following statement:
conda install -c oddt oddt
Scikit-learn It is a toolkit for predictive data analysis. To install in Conda execute the following statement:
conda install -c anaconda scikit-learn
Joblib It is a suite to provide lightweight pipelining in Python. To install in Conda execute the following statement:
conda install -c anaconda joblib
Corresponds to software not found as an Anaconda package.
MGLTools was developed in the Sanner lab at the Center for Computational Structural Biology (CCRB) and visualization and analysis of molecular structures. The installer can be downloaded from https://ccsb.scripps.edu/mgltools/downloads/
The software vina4dv is a fork of AutoDock Vina modified to output features for DeltaVinaRF20 score. The repository can be downloaded from https://github.com/chengwang88/vina4dv
For both MGLTools and vina4dv are necessary to create environment variables in .bashrc (Linux) or .bash_profile (macOS) file in the home directory:
# Set MGLTools-1.5.6
export PATH=$PATH:/home/myuser/MGLTools-1.5.6/bin
export MGL=/home/myuser/MGLTools-1.5.6/
export MGLPY=$MGL/bin/python
export MGLUTIL=$MGL/MGLToolsPckgs/AutoDockTools/Utilities24/
# Set vina4v
export VINADIR=/home/myuser/vina4dv/build/linux/release/
The DSSP (hydrogen bond estimation algorithm) program was designed by Wolfgang Kabsch and Chris Sander and calculates the secondary structure assignment given the 3D structure of a protein. To install the DSSP run the following statement:
sudo apt-get install -y dssp
The Java Runtime Environment (JRE) is a software layer to run Java programs. PaDEL-Descriptor is a program written in Java, it is inside the implementation of the scoring function and it is used to calculate ligand features. To install the JRE run the following statement:
sudo apt install default-jre
R is an integrated suite of software facilities for data science. The scoring function developed for this work was trained with R. To install the R run the following statement:
sudo apt install r-base
The machine learning Random Forest algorithm was used to train RFL-Score, therefore it is necessary to install its corresponding package within R. In the R console execute the following statement:
install.packages("randomForest")
The current repository includes the SF and features generation scripts.
To calculate the score of a protein-ligand complex.
Params:
- -r: Protein's file path. PDB format is mandatory.
- -l: Ligand's file path. MOL2 format is mandatory.
- -o: Output path.
- -n: Output file name.
- -g: Protein-ligand list file path.
- -t: Temporary folder path.
A score example:
python sf160.py -r examples/1a30/1a30_protein.pdb -f examples/1a30/1A30.fasta.txt -l examples/1a30/1a30_ligand.mol2 -n 1a30 -o /home/oarrua/Desktop/Test -t /home/oarrua/Desktop/Test
A protein-ligand list example:
python sf160.py -g examples/list_example -n all -o /home/oarrua/Desktop/Test
To calculate a set of features of a protein-ligand complex.
Features Set
| Set | Description |
|---|---|
| amino20 | Related features of amino acid percentages. |
| dssp34 | Related features of the secondary structure of proteins. |
| nn350 | NNscore 2.0 features |
| pd92 | A set ligand features of PaDEL-Descriptors. |
| rdkt2d147 | A set of 2D ligand descriptors of RDKit. |
| rdkt3d11 | A set of 3D ligand descriptors of RDKit. |
| sasa10 | Related features of solvent-accessible surface area. |
| vina58 | Vina implemented terms. |
| deltavina20 | Autodock Vina score. |
Params:
- -r: Protein's file path. PDB format is mandatory.
- -f: Protein Sequence's file path. FASTA format is mandatory.
- -l: Ligand's file path. MOL2 format is mandatory.
- -o: Output path.
- -n: Output file name.
- -g: Protein-ligand list file path.
- -t: Temporary folder path.
A simple example:
python sf723_descriptors.py -r examples/1a30/1a30_protein.pdb -f examples/1a30/1A30.fasta.txt -l examples/1a30/1a30_ligand.mol2 -n 1a30 -o /home/oarrua/Desktop/Test -x "amino20 dssp34 nn350 pd92 rdkt2d147 rdkt3d11 sasa10 vina58 deltavina20"
A protein-ligand list example:
python sf723_descriptors.py -g examples/list_example_with_amino20 -n all -o /home/oarrua/Desktop/Test -x "amino20 dssp34 nn350 pd92 rdkt2d147 rdkt3d11 sasa10 vina58 deltavina20"
RFL-SCORE Installation Guide: https://www.youtube.com/watch?v=lH0dlUFCz8Y
- Supporting for DSSP 4.0.4 (04-27-2024)
Arrua, O. E., Aderhold, A., Werhli, A. V., & Machado, K. D. S. (2024, August). RFL-Score: Random Forest with Lasso Scoring Function for Protein-Ligand Molecular Docking. In 2024 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB) (pp. 1-8). IEEE.
RFL-Score: Random Forest with Lasso Scoring Function for Protein-Ligand Molecular Docking