Challenges and progress with molecular dynamics simulation of disordered peptides

Abstract

Molecular dynamics (MD) simulations are a widely-accepted method for studying biomolecular systems. They are particularly useful for investigating stable protein structures where the range of configurations to be sampled is relatively small. Proteins and peptides with regions that lack regular structure pose a much more challenging sampling problem for MD simulations. An understanding of the structure and behaviour of such disordered regions is essential, because such regions are found in a significant number of proteins. Further, the lack of structure normally confers functional advantages, such as the capacity to bind several different other molecules. The need for reliable MD simulation methods is more important in such disordered cases, as experimental techniques are often inadequate. This thesis investigates the theory and practice of MD simulations, with a view to enhancing their capacity to simulate such systems efficiently on high-performance computers. In particular, the thesis examines in detail the replica-exchange molecular dynamics (REMD) method, which has become a popular tool for increasing sampling of challenging free-energy landscapes. It presents guidelines for best practice in running REMD simulations, and demonstrates a flaw in the theory that determines how replica exchanges occur. This flaw is shown to be significant in practice, and a solution is proposed and tested. The particle-mesh Ewald (PME) algorithm is widely used to treat the important electrostatic component of MD simulation models. This thesis shows how best to choose the parameters of the PME method in the MD simulation software GROMACS so that the full periodic Coulomb interaction is approximated at a known level of accuracy and minimal computational cost on various hardware. It is hoped that this will stimulate future work on the level of accuracy required for effective sampling. Also in this work, the performance of the inner loops of GROMACS on the BlueGene/L and BlueGene/P supercomputer families was improved by around 11%. Most of the aforementioned work has already appeared in peer-reviewed literature. Finally, these techniques were applied to large-scale microsecond-length REMD simulations of a disordered peptide. The peptide of interest was the tandem octapeptide repeat region of the N-terminal disordered region of human prion protein. Three of the tandem repeats were simulated. The results were compared with existing experimental studies of this peptide system. Some evidence of ordered structure was identified; in particular, interactions at close distance between side chains of all pairs of aromatic residues were observed. Show more