Integrating model-based design of experiments and computer-aided solvent design

Computer-aided molecular design (CAMD) methods can be used to generate promising solvents with enhanced reaction kinetics, given a reliable model of solvent effects on reaction rates. Herein, we use a surrogate model parameterised from computer experiments, more specifically, quantum-mechanical (QM) data on rate constants. The choice of solvents in which these computer experiments are performed is critical, considering the cost and difficulty of these QM calculations. We investigate the use of model-based design of experiments (MBDoE) to identify an information-rich solvent set and integrate this within a QM-CAMD framework. We find it beneficial to consider a wide range of solvents in designing the solvent set, using group contribution techniques to predict missing solvent properties. We demonstrate, via three case studies, that the use of MBDoE yields surrogate models with good statistics and leads to the identification of solvents with enhanced predicted performance with few iterations and at low computational cost.

Automated summary

Computer-aided molecular design (CAMD) methods are used to generate promising solvents with enhanced reaction kinetics by utilizing a reliable model of solvent effects on reaction rates. In this study, a surrogate model parameterized from computer experiments, specifically quantum-mechanical (QM) data on rate constants, is employed. The selection of solvents for these computer experiments plays a crucial role due to the cost and difficulty of QM calculations. Model-based design of experiments (MBDoE) is used to identify an information-rich solvent set and incorporate it into a QM-CAMD framework. The results of three case studies demonstrate that MBDoE can yield surrogate models with good statistics and identify solvents with enhanced predicted performance in a few iterations and at low computational cost.