Computational molecular design (CMD) is a methodology which applies optimization techniques to develop novel lead compounds for a variety of applications. The product design framework developed in this work seeks to accelerate the commonly used experimental trial-and-error approach by searching through large molecular spaces to provide a set of chemical structures likely to match a set of desired property targets. In this presentation, an overview of CMD methods used in our group is presented. Two major challenges are defined: the prediction of physical, chemical and biological properties of various molecular systems, and the determination of chemical structures matching a set of property targets within a large molecular space. To predict the physical and chemical properties of a specific class of molecules, quantitative structure-property relations (QSPRs) are developed which predict values of such properties as solubility, diffusivity, toxicity, polymer glass transition temperature, critical properties, and melting and decomposition temperatures. The selection of molecular descriptors for the QSPRs is performed using Mallow’s Cp statistic, which combines a goodness-of-fit score with a penalty for complexity. Prediction intervals are developed for errors in the fitted model as well as errors in property predictions for novel designs. The resulting property prediction models are then integrated within a computational molecular design framework, which combines the QSPRs with structural feasibility constraints in a combinatorial optimization problem. This problem is solved using a stochastic algorithm, Tabu Search, and 95% confidence intervals are computed for each solution. Two example systems are described in this presentation: the design of novel ionic liquids (ILs) for use as refrigerants and solvents, and the development of excipients for stabilization of protein drugs.
Copyright. All Rights Reserved.