A fast method that can predict the binding affinities of chemicals to a target protein with a high degree of accuracy will be very useful in drug design and regulatory science. We have been developing a scoring function for affinity prediction, which can be applied to extensive protein systems, and also trying to generate a prediction scheme that specializes in each target protein, with as high a predictive power as possible. In this study, we have constructed a prediction scheme with target-specific scores for estimating ligand-binding affinities to human estrogen receptor α (ERα), considering the major conformational change between agonist-and antagonist-bound forms and the change in protonation states of histidine at the ligand-binding site. The generated scheme calibrated with fewer training compounds (23 for the agonist-bound form, 17 for the antagonist-bound form) demonstrated good predictive power (a predictive r 2 of 0.83 for 154 validation compounds); this was also true for compounds with frameworks that were quite different from those of the training compounds. Our prediction scheme will be useful in drug development targeting ERα and in primary screening of endocrine disruptors, and provides a successful method of affinity prediction considering the major conformational changes in a protein.Key words binding affinity; prediction score; docking; protein conformational change; protonatioin state; estrogen receptorIn current drug discovery research, methods of searching for promising seed structures efficiently and optimizing them rationally decide the outcome of pharmaceutical development. The limitations of the high-throughput screening (HTS) technique are becoming clear, and accordingly, rational approaches to seed development by means of computational chemistry have been receiving attention again. In particular, in silico drug design based on the three-dimensional (3D) structure of a target protein has been enthusiastically attempted.1-3) The docking technique, which is the core technique of structurebased drug design (SBDD), has improved considerably over nearly three decades, 4,5) although difficult problems such as protein flexibility and consideration of water molecules remain. Meanwhile, another important technique, a prediction score for the binding affinities of small compounds to a target protein, 6) has not yet met the requirements for accuracy and computational cost.
4,7)One approach for predicting binding affinity is to try to estimate the binding free energy rigorously, e.g., the free energy perturbation (FEP) technique, 8) linear interaction energy method, 9) and Poisson-Boltzmann equation methods. 10,11) However, these rigorous methods have a large computational cost for every ligand-protein complex and would not be suitable for sequential evaluation of a number of compounds. Moreover, the FEP method can be adopted only to compare energies between ligands with the same structural frames. A number of empirical scores have also been developed. 4,12) These scores can handle many compounds ...