Binding Affinity Prediction for Protein-ligand Complexes #############################################################################

#############################################################################

1. Requirements BAP currently supports a Linux system and Python 3.7, and requires main dependency packages as follows.

   • deemchem (https://github.com/deepchem/deepchem)
   • rdkit (https://www.rdkit.org/)
   • numpy (https://numpy.org/)
   • pandas (https://pandas.pydata.org/)
   • sklearn (https://scikit-learn.org/stable/)
   • tensorflow (https://www.tensorflow.org/)
   • multiprocessing (https://docs.python.org/3/library/multiprocessing.html)
   • biopandas (http://rasbt.github.io/biopandas/)
   • scipy (https://www.scipy.org/)

2. Data downloading and preprocessing

   1. Downloading: Model construction - 'refined-set' data folder: PDBbind refined set v2020 (http://www.pdbbind.org.cn/) validation - 'casf2016' data folder: CASF-2016 set in PDBbind (http://www.pdbbind.org.cn/) - 'csarhiqS1' data folder: CSAR-HiQ sets 1 (http://www.csardock.org/) - 'csarhiqS2' data folder: CSAR-HiQ sets 2 (http://www.csardock.org/) - 'csarhiqS3' data folder: CSAR-HiQ sets 3 (http://www.csardock.org/)

   2. Preprocessing: 'refined-set': save the ligand files as PDB files and the protein files as MOL2 files (e.g. using software like UCSF Chimera) 'casf2016': save the ligand files as PDB files and the protein files as MOL2 files (e.g. using software like UCSF Chimera) 'csarhiqS1', 'csarhiqS2' and 'csarhiqS3': save the protein and ligand in each complex as PDB files (e.g. using software like UCSF Chimera), and name these files as those in PDBbind refined set (e.g. 1ax1_protein.pdb, 1ax1_ligand.pdb in '1ax1' folder)

   3. Put these folders together:

      • Create a root folder (e.g. 'PDBbind')
      • Create a 'v2020' foler in 'PDBbind', a 'PDBbind_v2020_refined' folder in 'v2020', and put 'refined-set' folder in 'PDBbind_v2020_refined'
      • Create a 'ValidationSets' folder, and put 'casf2016', 'csarhiqS1', 'csarhiqS2' and 'csarhiqS3' in 'ValidationSets'
      • Put the 'indexes' folder (provided in this repository) in 'PDBbind'

3. Example codes are provided in the 'Examples' folder in this repository

   1. IMCP-Score - Constructing an IMCP-based machine-learning SF on PDBbind refined set v2020 (excluding the validation sets) and validating it on the four validation sets (CASF-2016 and CSAR-HiQ sets) using Pearson's correlation and RMSE
   2. IMCPiDB-Score - Constructing an IMCPiDB-based deep-learning SF on PDBbind refined set v2020 (excluding the validation sets) and validating it on the four validation sets (CASF-2016 and CSAR-HiQ sets) using Pearson's correlation and RMSE