Mining PDB Subpockets to Rebuild Ligand Binding Conformations
We constituted a database of subpocket-fragment complexes (Fragmentor) through deconvolution of each PDB (Protein Data Bank) ligand in all possible fragments that match one of the chemical molecule contained in the Pubchem database. After application of a Matriochka filter, we obtained a set of 28.482 2D-fragments and 398.236 3D-fragments (PDB conformations). Subpockets were defined as protein surfaces located at 4,5 Å around fragments.
The goal of this work was to determine if there is enough information in the PDB to successfully rebuilt the binding mode of ligands, starting from the target protein structure. To test this hypothesis, we selected ligands in unique PDB entry to build a test set of 2292 protein-ligand complexes and tried to recover the position and conformation of the ligands using our Fragmentor database. We were able to predict at least 80% of the 3D-structure from 1091 ligands (48%). In conclusion, this study highlights the quality of the information contained in the PDB and supports the use of its structural information for docking tools or fragment-based drug design.
Click here to download the PDF poster