Skip to content

college-of-pharmacy-gachon-university/SMPLIP-Score

Repository files navigation

Overview

SMPLIP-Score

SMPLIP-Score was develped for prediction of absolute ligand-protein binding affinities.

image

Requirements:

The following necessary packages should be installed in to process, generate fingerprint, train and test your model.

Procedures:

  1. Process the protein PDB files using KNIME workflow (PDBbind_v2015_Preprocessing).

    => This will generate correct protein PDB files in *.mol2 file format.

    The interface of KNIME workflow to process the protein (PDB) file.

    image

    => To know, how to import the PDBbind_v2015_Preprocessin.knwf workflow into KNIME Analytics Platform visit this tutorial (official channel of KNIMETV).

    https://www.youtube.com/watch?v=4GiwmM-qcC4

  2. Generate the protein-ligand binding fingerprints using IChem program.

    # run the following command to generate *.ifp file
    
    IChem 1a30_protein.mol2 1a30_ligand.mol2 > 1a30.ifp
    
    
  3. Generate substructutal molecular ligand fragments using SMF program.

     - SMF program provides GUI.
    

    image

     - The user must provide all the ligands in *.sdf file format.
    
  4. Generate a matrix of fingerprint from the output file IChem program using KNIME workflow (IChem_IFP_PDBBind_2015)

    => This worflow, took input of all *.ifp files from the directory, extract all relevant information and save into *.csv format.

  5. Train, validate and test your RF and DNN model using python script.

    => To train, validate and test the RF model use the Train-Valid-Test-RF.ipynb [SMPLIP-Score]

     Load the *.ipynb* file using the jupyter notebook.
    

    => To train, validate and test the DNN model use the Train_Valid_Test-DNN.py [SMPLIP-DNN]

    # usage argument (bash shell):
    
    python Train_Valid_Test-DNN.py > log.txt
    
    # see the comment section of Train_Valid_Test-DNN.py for detailed information on file uses.
    
    

Additional Information:

For any queries mail to Mi-hyun Kim (kmh0515@gachon.ac.kr) or Surendra Kumar (surendramph@gmail.com)

References:

SMPLIP-Score: predicting ligand binding affinity from simple and interpretable on-the-fly interaction fingerprint pattern descriptors

HTML Manuscript