drug-design-dynamics

Installation

1. Clone the repository:

   git clone https://github.com/Simran-Sodhi/drug-design-dynamics.git
   

Project Structure

• PLAS20K/: Contains datasets related to the PLAS20K project
• Rshiny/: Scripts and resources for the R Shiny application
• SPICE/: Data and parsing scripts pertaining to the SPICE project // we are not using this data anymore
• machineLearningMethods/: Machine learning models and training scripts
• metropolisMethod/: Implementation of the Metropolis algorithm

To download input data for Metropolis simulation from PLAS20K

Input: extended_PLAS20K.csv

• usePDBnames.go to extract all the pdb_id of protein-ligand complex used in PLAS20K (stored in PLAS20K_pdb_ids.txt)
• use batch_download.sh to grab all the pdb files from RSCB
• use splitPDB.go to seperate proteins and ligands (output two pdb files for proteins and ligands)
• use convert_pdb_to_mol2.sh (calls Open Babel) to convert all pdb files to mol2 files.

Running the metropolis simulation from the go code

• You can use the metropolisMethod/main.go to run the metropolis simulation
• You need to provide data in metropolisMethod/Data. Some sample data is present there
• In main.go there are three options: one to simulate multiple ligands RunMultipleLigands(), one to get RMSD values: TestMethodRMSD() and the third for the R Shiny app: RShinyAppMain(args []string)
• All the outputs go into the metropolisMethod/Output folder

R shiny

Interactive web app for predicting protein-ligand interations by evaluating their binding energies using Metropolis and machine learning simulations method.

Requirements

• Set your Python path at Line 241 in app.R to ensure that the ML python script can be executed.
• Put all external data under folder either ./MCdata or ./MLdata for data to be accessible (like the example data).

Usage

• Metropolis:
  Tha app takes one protein file (.pdb/.mol2) and multiple ligand files (.mol2 files uploaded together in a directory) as input,
  after successfully uploaded the data, hit "Run Simulation" button, and you will get:
  
  (1) the plot of all protein-ligand pairs' binding energies
  (2) the structure of protein-ligand pair with minimum binding energy shown
  (need to preprocess the output protein and ligand file (.mol2) using Chimera or PyMOL to generate the video and save as ./output/results.mp4)
  
  

• Machine Learning:
  The app takes protein-ligand interaction dataset (.csv file) as input,
  you can then select features you are interested in from ["electrostatic", "polar_solvation", "non_polar_solvation", "vdW"],
  and select models you would like to use from ["RandomForest", "DecisionTree", "XGBoost", "LightGBM", "SVM"],
  after successfully uploaded the data and selected the parameters, hit "Run Simulation" button, and you will get:
  
  (1) the evalution matrix of all models (evalution method includes: MSE, RMSE, MAE, R2, MAPE)
  (2) feature importance tables generated by models you selected
  

Recorded Code demo available at:

https://drive.google.com/drive/folders/1g_GTiWV2_l0lUbO9OTQ9euYVeeyWyaBW?usp=drive_link