Physicochemical property prediction for small molecules using integral equation-based solvation models

Abstract

In this thesis the accurate prediction of physicochemical properties of small, pharmaceutically relevant compounds is investigated. To predict condensed phase properties such as hydration free energies, acid dissociation constants (pKa), and distribution and partition coefficients (log D and log P, respectively) it is necessary to accurately describe the solute, the solute-solvent interactions, and the solvent-response to the solute’s presence. When this is achieved, the Gibbs energies of the molecules in solution can be used to calculate macroscopic physicochemical properties. The embedded cluster reference interaction site model (EC-RISM) makes it possible to combine a quantum chemical description of the solute with an accurate solvent response via the three-dimensional reference interaction site model (3D RISM). This is ideal for calculating physicochemical properties of small molecules, because EC RISM yields both the electronic energy of the solvent-polarized wave function, as well as the excess chemical potential of the molecule in solution, the sum of which can be defined as the Gibbs energy of the molecule in solution. The development of solvent susceptibilities for the non-aqueous solvents cyclohexane and n octanol is reported, as well as the challenges and implications of including water saturation for organic solvents. The solvent susceptibilities are used to train partial molar volume corrections to correct for the error inherent in the calculation of the 3D RISM excess chemical potential using reference data from the Minnesota solvation database (MNSOL). Additionally, a method to calculate accurate pKa values is presented and the formal equivalence of a microstate transition and a partition function approach is briefly summarized. The performance of the models is benchmarked by participation in the Statistical Assessment of Modeling of Proteins and Ligands (SAMPL) challenges. First, the SAMPL5 challenge, where cyclohexane-water distribution coefficients log D7.4 had to be calculated. In subsequent challenges the task was split into determining aqueous pKa values during the SAMPL6 challenge and octanol-water partition coefficients log P of a subset of these compounds for SAMPL6 part II. Over the course of these challenges a number of key improvements were made to the EC RISM model, often directly as a result of inconsistencies or performance issues during one of the SAMPL challenges. Finally, an extension of the partial molar volume correction to extreme conditions such as high pressure is reported.