Otimização de protocolo computacional para predição de valores de pKa em solução de dimetilsulfóxido com modelos implícito e híbrido

Abstract

The acidity or basicity of a compound consists of a property that is a direct reflection of its propensity to donate or accept a proton and its knowledge is of fundamental importance for the understanding of several chemical and biochemical processes. This property can be expressed through the substance’s pKa. When the system of interest is modeled in solution, the results obtained are often not satisfactory as in the gas phase, as there are difficulties in measuring the impacts caused by the effect of the solvent acting on the reaction, as evaluated in some studies of physico-chemical properties of solvated systems Few studies have been carried out on the prediction of pKa in aprotic solvents such as dimethyl sulfoxide (DMSO). The present study proposes the optimization of a computational protocol in order to make prediction possible with absolute errors below 1.8 units of pKa in relation to the experimental values. The calculations in this work were performed at the level M06-2X/6-31+G(d) with the implicit solvation model C-PCM aiming at the prediction of pKa values in DMSO for 105 organic compounds not correlated through a direct procedure. Two steps were taken in this study. The first step consisted of calculating the free energies of the initial and final states of the deprotonation reaction of 105 acids in DMSO using only the C-PCM solvation model. The second stage of the study aimed to calculate the absolute pKa of the first 10 structures in a hybrid approach, which included an explicit DMSO molecule, treating the other solvent molecules implicitly with the continuous solvation model. The explicit DMSO molecules were positioned from the trajectory analysis of ab initio molecular dynamics simulations performed at the GFN2-xTB level and, subsequently, calculations were performed at the same level as the first stage. The results obtained, for the calculation of pKa only with implicit solvation, showed a significant correlation coefficient (r 2 = 0.950), indicating a high predictive power. In the hybrid approach it was possible to obtain an average absolute error slightly above those obtained by the implicit approach for the prediction of absolute values and significantly less than those obtained for the relative values.