Premium
Towards a Realistic Representation in Surface‐Based Pseudoreceptor Modeling: a PDB‐Wide Analysis of Binding Pockets
Author(s) -
Wilson Gregory L.,
Lill Markus A.
Publication year - 2012
Publication title -
molecular informatics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.481
H-Index - 68
eISSN - 1868-1751
pISSN - 1868-1743
DOI - 10.1002/minf.201100166
Subject(s) - overfitting , gaussian , representation (politics) , surface (topology) , quantitative structure–activity relationship , biological system , protein ligand , property (philosophy) , surface protein , computer science , chemistry , artificial intelligence , mathematics , computational chemistry , machine learning , geometry , artificial neural network , biology , philosophy , organic chemistry , epistemology , politics , political science , law , virology
Surface‐based pseudoreceptor methods are expansions of 3D‐QSAR techniques placing physico‐chemical information onto a 3D surface surrounding a set of aligned compounds that bind into the same binding site of a common protein target. With this mapping pseudoreceptor methods attempt to create models of the target protein binding site around the ligand ensemble. The surface points of the pseudoreceptor model are typically independent descriptors and property mapping onto the surface points is prone to overfitting. In this article, we developed surface descriptors based on 2D Gaussian functions that can model the physico‐chemical properties of binding sites of proteins. Binding pocket surfaces of a large set of experimentally determined protein‐ligand complex structures are analyzed and 2D Gaussian functions are used to fit the surface properties. The fitted property values differ from the original values on average by 15–25 %, and on average six Gaussian functions are necessary to model each surface property. These descriptors allow for a realistic representation of the binding site and will limit the number of descriptors used throughout the QSAR optimization phase.