Files
Abstract
While linear free energy relationships (LFERs) have been used for decades to relate electronic and steric features of a substituent to the reaction rate, and thus infer something about the reaction mechanism they have recently become more popular as tools for catalyst and ligand design. In these cases, multi-parameter fitting is used to create a function that predicts an experimental outcome - such as reaction rate, selectivity, binding affinity, or inhibitory concentration - based on computationally determined descriptors of the ligand or catalyst. However, these fits are often complex and difficult to interpret. I have proposed and developed a series of multivariate LFERs that use purely computational data. This allows us to isolate features of the complex chemical system under consideration, and thus achieve both quantitative prediction of interaction energetics and clear interpretability, providing chemical insight into the underlying interactions that influence the observed trends. I have applied this methodology to both the areas of catalyst design and drug optimization. In the area of catalyst design I was able to propose a new catalyst for a BINOL-catalyzed asymmetric conjugate addition which has since been experimentally shown to have improved yield. In the area of drug design and optimization I have developed new descriptors of heterocycle electrostatics that have been successfully used to describe both traditional and non-traditional stacking interactions that, while observed in drug binding sites, had previously not been well understood or fully taken advantage of. Further, I have developed methodology for the rapid evaluation of these descriptors from atom connectivity information such as a SMILES string, eliminating the need for expensive quantum mechanical computations to make accurate predictions.