Theoretical and computational coarse-grained models for the study of molecular systems.

摘要

This dissertation presents novel theoretical and computational methods for performing large-scale simulations of molecular systems. In spite of dramatic advances in computational technology, the most extensive all-atom protein simulations are currently limited to the microsecond timescale and can only be applied to relatively small systems. This work addresses that problem by developing a variety of novel theoretical and computational models that treat protein residues as unified interaction centers. The advantages and limitations of these reduced models are analyzed.; A first goal of this work is to develop a minimal theory of the kinetics and equilibrium properties of proteins, applicable to the helix-coil transition in peptides and to the general case of protein folding. Starting from an Ising model for peptide energetics, the equilibrium properties of the helix-coil transition and the essential features of its kinetics are explained. A new method for estimating the mean first passage time for this transition is developed and its results are used to interpret the results of recent experimental studies.; The second goal is to develop and test a novel class of inter-residue potentials for proteins, dependent on the distance between the interaction centers and on relative residue-residue orientations. Statistical analysis methods are employed to derive parameters for the interaction potentials from protein structural databases. The potentials are validated by native state identification tests using databases of protein decoy sets. A new eigenvalue analysis method, which permits the ranking of amino acids according to their relative contributions to the global features of the potential, is presented and used to make objective comparisons of different interaction schemes.; The third aim is to model molecular fluids using coarse-grained models. By extending our methodology used to derive reduced models for proteins, a novel statistical potential for water is derived from large-scale molecular simulations of liquid water. The resulting model is tested and shown to reproduce essential features of water structure.; These novel methods are useful for improving the quality and the performance of coarse-grained, off-lattice protein simulations, and for providing a better understanding of the mechanisms that govern the structure and the kinetics of protein systems.