Overview

SMPLIP-Score

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

Requirements:

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

• KNIME analytics platform (> 3.5)
• Community nodes and schrodinger nodes should be installed (For schrodinger node please follow this link: https://www.schrodinger.com/kb/1085)
• IChem program (http://bioinfo-pharma.u-strasbg.fr/labwebsite/download.html) [License can be obtained upon request to Dr. Didier Rognan]
• SMF program (http://vpsolovev.ru/programs/smf/)
• python >3.0
• Keras (2.3.1)
• Tensorflow (2.1.0)
• Scikit Learn (0.22)
• Jupyter Notebook

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.

   

   => 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.
   

   

    - 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 ([email protected]) or Surendra Kumar ([email protected])

References:

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