How well does a restrained electrostatic potential (RESP) model perform in calculating conformational energies of organic and biological molecules?

In this study, we present conformational energies for a molecular mechanical model (Parm99) developed for organic and biological molecules using the restrained electrostatic potential (RESP) approach to derive the partial charges. This approach uses the simple “generic” force field model (Parm94), and attempts to add a minimal number of extra Fourier components to the torsional energies, but doing so only when there is a physical justification. The results are quite encouraging, not only for the 34-molecule set that has been studied by both the highest level ab initiomodel (GVB/LMP2) and experiment, but also for the 55-molecule set for which high-quality experimental data are available. Considering the 55 molecules studied by all the force field models for which there are experimental data, the average absolute errors (AAEs) are 0.28 (this model), 0.52 (MM3), 0.57 (CHARMm [MSI]), and 0.43 kcal/mol (MMFF). For the 34-molecule set, the AAEs of this model versus experiment and ab initio are 0.28 and 0.27 kcal/mol, respectively. This is a lower error than found with MM3 and CHARMm, and is comparable to that found with MMFF (0.31 and 0.22 kcal/mol). We also present two examples of how well the torsional parameters are transferred from the training set to the test set. The absolute errors of molecules in the test set are only slightly larger than in the training set (differences of <0.1 kcal/mol). Therefore, it can be concluded that a simple “generic” force field with a limited number of specific torsional parameters can describe intra- and intermolecular interactions, although all comparison molecules were selected from our 82-compound training set. We also show how this effective two-body model can be extended for use with a nonadditive force field (NAFF), both with and without lone pairs. Without changing the torsional parameters, the use of more accurate charges and polarization leads to an increase in average absolute error compared with experiment, but adjustment of the parameters restores the level of agreement found with the additive model. After reoptimizing the Ψ, Φ torsional parameters in peptides using alanine dipeptide (6 conformational pairs) and alanine tetrapeptide (11 conformational pairs), the new model gives better energies than the Cornell et al. ( J Am Chem Soc 1995, 117, 5179–5197) force field. The average absolute error of this model for high-level ab initio calculation is 0.82 kcal/mol for alanine dipeptide and tetrapeptide as compared with 1.80 kcal/mol for the Cornell et al. model. For nucleosides, the new model also gives improved energies compared with the Cornell et al. model. To optimize force field parameters, we developed a program called parmscan, which can iteratively scan the torsional parameters in a systematic manner and finally obtain the best torsional potentials. Besides the organic molecules in our test set, parmscan was also successful in optimizing the Ψ, Φ torsional parameters in peptides to significantly improve agreement between molecular mechanical and high-level ab initio energies. © 2000 John Wiley & Sons, Inc. J Comput Chem 21: 1049–1074, 2000