Electrostatic evaluation of isosteric analogues

A method is presented for enumerating a large number of isosteric analogues of a ligand from a known protein-ligand complex structure and then rapidly calculating an estimate of their binding energies. This approach takes full advantage of the observed crystal structure, by reusing the atomic co-ordinates determined experimentally for one ligand, to approximate those of similar compounds that have approximately the same shape. By assuming that compounds with similar shapes adopt similar binding poses, and that entropic and protein flexibility effects are approximately constant across such an isosteric series ("the frozen ligand approximation"), it is possible to order their binding affinities relatively accurately. Additionally, the constraint that the atomic coordinates are invariant allows for a dramatic simplification in the Poisson-Boltzmann method used to calculation the electrostatic component of the binding energy. This algorithmic improvement allows for the calculation of tens of thousands of binding energies per second for drug-like molecules, enabling this technique to be used in screening large virtual libraries of isosteric analogues. Most significantly, this procedure is shown to be able to reproduce SAR effects of subtle medicinal chemistry substitutions. Finally, this paper reports the results of the proposed methodology on seven model systems; dihydrofolate reductase, Lck kinase, ribosome inactivating protein, L: -arabinose binding protein, neuraminidase, HIV-1 reverse transcriptase and COX-2.