Iuzzolino, Luca; (2018) Expanding crystal structure prediction to larger and more flexible molecules of pharmaceutical interest. Doctoral thesis (Eng.D), UCL (University College London).

Preview

Text Final thesis Luca Iuzzolino.pdf - Accepted Version Download (13MB) | Preview

Abstract

The use of Crystal Structure Prediction (CSP) studies in the pharmaceutical industry is currently limited by computational cost, which scales badly with molecular size and flexibility. This thesis seeks to develop new methods that would allow to perform CSP studies on larger, more flexible pharmaceutical-like molecules. First, a full CSP workflow was successfully used to predict the crystal structure of a large flexible molecule for the 6th Blind Test and in a joint computational-experimental study of the antihelminthic drug mebendazole. These CSP studies were integrated with three previously published computational analyses of flexible pharmaceuticals and used to benchmark the development of new methods. Successively, knowledge-based conformational information retrieved from the Cambridge Structural Database (CSD) was used to facilitate the generation of candidate crystal structures of these five molecules. Millions of crystal structures were generated at a reduced computational cost, but with an equally effective coverage of the conformational search space, compared to the original CSP efforts. The importance of treating conformational flexibility when optimising search-generated crystal structures was then assessed. This led to using dispersion-corrected density functional tight-binding (DFTB-D) as an intermediate step to minimise all intra- and intermolecular degrees of freedom of several thousands of search-generated crystal structures. DFTB-D reduced the cost of the final lattice energy evaluations by providing better starting points, and results of similar quality to the original CSP studies were obtained after optimising only the intermolecular interactions with a higher quality wave-function. Finally, a CSD survey was performed to determine thresholds that can discriminate the great majority of polymorphs from duplicate determinations. These thresholds and comparison methods were implemented in a Python programme that can be used in CSP studies to perform clustering and to interpret the results more effectively. The prospects for expanding the use of CSP to pharmaceutical development are discussed.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item