Desenvolvimento e aplicação de protocolos de solvatação baseados em uma abordagem híbrida multiescala sequencial

Abstract

Describing the solvated phase remains a significant challenge, especially when the system contains protic and polar solvents, such as water, directly impacting the accuracy in predicting the properties of interest. This work aims to develop protocols that account for the effect of explicit solvation in the calculation of thermodynamic properties (especially Gibbs’ free energies), with the goal of reproducing experimental data within chemical accuracy, defined as an absolute error below 1 kcal mol-1. To achieve this, a multiscale hybrid approach was employed, characterized by the combination of simulation methods at different resolution levels: quantum mechanics, molecular mechanics, and implicit solvation models. As an evaluation metric, both theoretical and experimental reduction potentials of a set of redox reactions were considered. Although conceptually simple, modeling reduction potentials presents high operational complexity, particularly when aiming to achieve chemical accuracy. The methodological development involved the following steps: (i) performing a benchmark of reduction potentials in water using 25 density functionals; (ii) inclusion of explicit water molecules forming hydrogen bonds with the solute, incorporated into the reduction potential calculations; and (iii) inclusion of explicit water molecules in a solute-solvent ratio corresponding to the standard concentration (1 mol kg 1). The study − enabled the development of tools capable of automatically selecting solvent molecules in the vicinity of the modeled systems, positioned through classical molecular dynamics simulation procedures. These applications compose the suite named Solvate Suite. The results showed absolute mean errors ranging from 4.78 kcal mol-1 to 27.98 kcal mol-1 for the reduction potentials. However, this same methodology was also applied to the calculation of chemical shifts of 13C for a set of molecules from different categories, for which mean errors below 1.5 ppm were obtained. This performance enabled a more accurate prediction of the order of the peaks in the NMR spectrum of two isomeric carbohydrates, overcoming several incorrect assignments associated with the use of the implicit solvation model C-PCM.