On the Strengths and Limitations of an Automated Docking Procedure: Application to Rigid and Highly Flexible Ligands

The prediction of a protein-ligand interaction when the interaction site is known is a relatively simple task. The greater challenge is to be able to predict both the mode and
location of the interaction, preferably in the absence of any information other than the structures of the molecules themselves.

A 'blind' docking approach, using an iterative genetic algorithm, challenged large areas of the target proteins in order to determine the mode and location of ligand
interaction. Progressively complex systems were analysed commencing with small and large rigid ligands, followed by small flexible ligands and finally complex flexible ligands interacting with a viral capsid protein that lacked a classically defined binding site. Manual screening methods and a post-processing algorithm were developed for the examination of the output from flexible ligand docking simulations.
A database of annotated three-dimensional structures was also produced in an effort to identify potential interaction sites.

Both the mode and location of interactions involving rigid ligands were predicted with a high degree of confidence. Interactions involving flexible ligands were predicted
with a confidence that was inversely proportional to ligand flexibility, to the extent that interaction sites were initially unidentifiable in the simulations examining greatest flexibility. Application of the post-processing algorithm identified flexible ligand binding sites with good correlation to experimentally determined results.

Recent developments in automated docking procedures have lessened the impact of ligand flexibility when the intended binding site is known, but it becomes of paramount importance when attempting to determine the location of an interaction on a protein surface. Supplementary information will often be required to identify binding sites that are not classically defined by cavities or depressions on the protein surface. The development of a post-processing algorithm and an annotated structure database will help to identify these sites and allow for more focussed interaction
prediction.

This thesis supplied via ROAR to UEL-registered users is protected by copyright and other intellectual property rights, and duplication of any part of the material is not permitted, except for your personal use for the purposes of non-commercial research and private study in electronic or print form. You must obtain permission from the copyright-holder for any other use. Electronic or print copies may not be offered, for sale or otherwise, to anyone. No quotation from the thesis may be published without proper acknowledgement.