New refinement strategies for a pseudoatom databank – toward rapid electrostatic interaction energy estimations

Abstract

Pseudoatom databanks, collections of parameters from the multipole model of electron densities for various atom types, are used to replace the Independent Atom Model with the more accurate Transferable Aspherical Atom Model (TAAM) in crystal structure refinements. The databanks are also employed to reconstruct the electron density of a molecule, crystal or biomacromolecular complex in a fast yet accurate way and compute various properties such as the energy of electrostatic interactions, for example. A even faster but similarly accurate model for estimations of electrostatic energy exists called aug-PROmol [Bojarowski, Kumar & Dominiak (2016). ChemPhysChem, 17, 2455–2460]. A model analogous to aug-PROmol cannot be built from the current pseudoatom databanks, as they perform badly when truncated to the monopole level. Here, new strategies for multipole model refinements were sought, leading to better parametrization at the monopole level. This would allow the creation of a pseudoatom databank in a single route of model parametrization, which would be suitable for both crystal structure refinement and rapid electrostatic energy calculations. Here it is shown that the cumulative approach to multipole model refinements, as opposed to simultaneous or iterative refinements of all multipole model parameters (Pv , κ, Plm , κ′), leads to substantially different models of electron density. Cumulative refinement of two blocks of parameters, the first with Pv and κ and then the second with Plm and κ′, leads to the P v κ|P lm κ′ model having promising properties. The P v κ|P lm κ′ model is as good as the University at Buffalo DataBank (UBDB) in X-ray structure TAAM refinements and electrostatic energy estimations, especially for less polar molecules. When truncated to the monopole level, the P v κ model has a chance to replace aug-PROmol in fast yet accurate electrostatics energy calculations, although some improvements in κ parametrization for polar functional groups are still needed. The P v κ model is also a source of point charges which behave similarly to restrained electrostatic potential (RESP) charges in electrostatic interaction energy estimations.