| Summary: | <p>The generation of conformations for small molecules is one of the cornerstones of computational chemistry. Identifying diverse low energy conformers and, in particular, the lowest energy conformer are essential for
applications, such as molecular docking and molecular property predictions. The large conformational space of flexible molecules and the high computational cost of accurate energy evaluation with methods such as
quantum mechanics are two key challenges. </p>
<p>This thesis explores the use of statistical techniques to (i) understand the factors governing the conformational preferences of small molecules and their population, and (ii) improve the efficiency in finding the lowest energy conformer of a molecule. I first provide an overview of conformer sampling and Bayesian optimisation, followed by an introduction to circular data analysis for analyzing torsional distribution in small molecule conformations. </p>
<p>I demonstrate the effectiveness of Bayesian optimisation algorithm in finding the lowest energy conformation of molecules, which requires orders of magnitude fewer energy evaluation to find the lowest energy conformation. I also show how sampling efficiency can be further improved by biasing the search towards low energy regions through a knowledge-based acquisition function. To extend the sampling framework, I explore the use of Cremer Pople puckering parameters to characterise complex ring geometries, and study the resulting ring puckering preferences extensively. </p>
<p>Finally, I investigate the factors contributing to the conformational entropies of small molecules, and develop linear models that predict the conformational entropies of small molecules accurately and rapidly. </p>
<p>In summary, this thesis contributes an improved understanding of small molecule conformational preferences, and introduces new methods to improve the efficiency of sampling conformers and entropy calculations.</p>
|
|---|