Journal of Computer-Aided Molecular Design

, Volume 13, Issue 3, pp 243-258

First online:

A comparative study of ligand-receptor complex binding affinity prediction methods based on glycogen phosphorylase inhibitors

  • Sung-Sau SoAffiliated withDepartment of Chemistry and Chemical Biology, Harvard University
  • , Martin KarplusAffiliated withDepartment of Chemistry and Chemical Biology, Harvard University

Rent the article at a discount

Rent now

* Final gross prices may vary according to local VAT.

Get Access


Finding an accurate method for estimating the affinity of protein ligands activity is the most challenging task in computer-aided molecular design. In this study we investigate and compare seven different prediction methods for a set of 30 glycogen phosphorylase (GP) inhibitors with known crystal structures. Five of the methods involve quantitative structure-activity relationships (QSAR) based on the 2D or 3D structures of the GP ligands alone. They are hologram QSAR (HQSAR), receptor surface model (RSM), comparative molecular field analysis (CoMFA), and applications of genetic neural network to similarity matrix (SM/GNN) or conventional descriptors (C2GNN). All five QSAR-based models have good predictivity and yield q2 values ranging from 0.60 to 0.82. The other two methods, LUDI and a structure-based binding energy predictor (SBEP) system, make use of the structures of the ligand-receptor complexes. The weak correlation between biological activities and the LUDI scores of this set of inhibitors suggests that the LUDI scoring function, by itself, may not be a general method for reliable ranking of a congeneric series of compounds. The SBEP system is derived from a set of physical properties that characterizes ligand-receptor interactions. The final neural network model, which yields a q2 value of 0.60, employs four descriptors. A jury method that combines the predictions of the five QSAR-based models leads to an increase in predictivity. A multi-layer scoring system that utilizes all seven prediction methods is proposed for the evaluation of novel GP ligands.

binding affinity prediction CoMFA genetic neural network glycogen phosphorylase inhibitor QSAR structure-based drug design