deexposito/bs_predictor

Machine Learning-Based Binding Site Prediction for Protein Structures

BS_predictor: Machine Learning-Based Binding Site Prediction for Protein Structures

---

BS_predictor is a supervised learning method designed to predict binding site residues within a protein structure. Leveraging machine learning techniques, the model analyzes the properties and interactions of residues to make accurate predictions. This Python program is tailored for UNIX-like environments.

Authors:

• Denis Expósito Navarro ([email protected])
• Marta Alonso Caubilla ([email protected])
• Yolanda Andrés López ([email protected])

Version: 2.0.0 (December 2023)

• Fix installation issues (setup.py)
• Provide a script for previous data preparation (+ option to modify it and extract different training files) and the binding site residue list file generated as csv.
• Provide a script for the generation of the training feature matrix and the feature matrix file as csv (feature_matrix.py)
• Provide script to train the model with the feature matrix with GridSearch and save the best model to use it in the main script (BSmodel.pkl)

Pre-requisites

The model requires the following packages:

Package	Version
pandas	>= 2.0
biopython	>= 1.81
scikit-learn	>= 1.3.2
numpy	>= 1.23.4
networkx	>= 3.1
scipy	>= 1.10.1
joblib	>= 1.3.2

To install these packages, run:

pip install -r requirements.txt

Additionally, you need to install the DSSP program to access the DSSP database of protein secondary structures. Use the following commands in a UNIX-like environment:

sudo apt-get install dssp
sudo ln -s /usr/bin/mkdssp /usr/bin/dssp

Installation

To install BS_predictor, navigate to the project folder containing setup.py and run:

python setup.py install

This installs the program BS_predictor.py, making the command BS_predictor available in the command line.

To make BS_predictor executable from any location in the command line, you can add its directory to your system's PATH. Add the following line to your shell configuration file (e.g., ~/.bashrc or ~/.zshrc):

export PATH="/path/to/BS_predictor:$PATH"

Replace "/path/to/BS_predictor" with the actual path to the directory containing the BS_predictor.py script. Then, restart your terminal or run:

source ~/.bashrc

Usage

Run BS_predictor in the command line using the following syntax:

BS_predictor -p <input_PDB_file> -o <output_PDB_file_name>

Output

The program generates:

• List of residues that belong to the predicted binding site.
• PDB file with information for residues in the predicted binding site to be represented in Chimera or Pymol.

Example

For the protein structure 4ins.pdb (blue) the following residues in the predicted binding site (ywllow) are obtained:

• 4ins_A_GLY_1
• 4ins_B_GLY_23
• 4ins_D_GLY_23