Abstract/Summary

Introduction Proteins form the very basis of life and are known to regulate a variety of activities within living organisms. Exploration on protein structure and function will be beneficial for the drug industry, food production and combating infectious disease. For the last few decades, solving protein structure has been dependent on experimental techniques that are very accurate such as nuclear magnetic resonance (NMR) and X-ray crystallography. However, recent advances in large-scale sequencing technologies, such as next-generation sequencing has increased the gap between the known protein sequence and solved protein structures. Therefore, devising effective computer programs is the only option available to help close the gap. At the moment, the most accurate protein prediction protocol is the comparative modelling method which consists of several steps such as template recognition, alignment, quality assessment and refinement. Among these steps, protein structure refinement remains a mostly unsolved problem as there are only a few methods that can consistently and efficiently improve over the starting protein model. In every molecular biology application, the quality of any predicted protein structure is of paramount importance, and the refinement process is the key to achieving higher quality 3D models. Here, we aim to develop a new hybrid refinement approach that can specifically assigned to works on refinement targets. Therefore in this study, we investigate the potential of the classical (ModFOLD5_single, ModFOLDclut2, ProQ2) and the new generation (ModFOLD6 _rank, ModFOLD6_global , ModFOLD6_cor, ProQ2D, ProQ3D, ProQRosCen, ProQRosCenD, ProQRosFA and ProQRosFAD) Accuracy Self Estimates (ASE) programs to be used to guide the iterative and Molecular Dynamic (MD) refinement process, using state of the art refinement tools such as i3Drefine and NAno Molecular Dynamic simulation (NAMD ). The performance of the newly developed protocol in each chapters such as ReFOLDa , ReFOLDb and ReFOLD were measured and analysed using a wide variety range of metrics to interpret the results. The development of new hybrid refinement approach such as ReFOLD has boost our group performance and helped us to gained second place for the Template Based Modelling (TBM) +TBM/ Free Modelling (FM) category of the tertiary structure prediction category in the recent CASPI 2 experiment. However the performance analysis of ReFOLD clearly indicate that another specialist refinement programs is needed as ReFOLD only managed to improve some of the targets from the more difficult refinement category . Thus the final working chapter of this thesis we focused on the development of two new hybrid refinement approaches (ReFOLDc and ReFOLDd) that were specifically assigned to work with the refinement targets. Among the two new developed refinement protocol, it is clear that the combination between ReFOLDa and ReFOLD have been shown to add value to the ReFOLDd pipeline when each of the protocols were employed selectively based on the quality of the starting models. Thus, we firmly believe that standard ReFOLD server approach should only be used to refine the already low­ quality initial models while ReFOLDb should be used on high-quality initial models. Together the combination of both protocols (ReFOLDd) will help us to bring 3D models of proteins closer to native structures and is therefore a helpful step towards closing the knowledge gap between protein structures and protein sequences. The author recommends further modification to ReFOLDd to further improve its refinement ability.