Understanding protein structure and dynamics: From comparative modeling point of view to dynamical perspectives

In this thesis, we have advanced a set of distinct bioinformatic and computational tools to address the structure and function of proteins. Using data mining of the protein data bank (PDB), we have collected statistics connecting the propensity between the protein sequence and the secondary structure. This new tool has enabled us to evaluate new structures as well as a family of structures. A comparison of the wild type staphylococcal nuclease to various mutants using the proposed tool has indicated long-range conformational deviations spatially distant from the mutation point. The energetics of protein unfolding has been studied in terms of the forces observed in molecular dynamics simulations. An adaptive integration of the steered molecular dynamics is proposed to reduce ground state dominance by the rare low energy trajectories on the estimated free energy profile. The proposed adaptive algorithm is utilized to reproduce the potential of mean force of the stretching of decaalanine in vacuum at lower computational cost It is then used to construct the potential of mean force of this transition in solvent for the first time and to observe the hydration effect on the helix-coil transformation. Adaptive steered molecular dynamics is also implemented to obtain the free energy change during the unfolding of neuropeptide Y and to confirm that the monomeric form of neuropeptide Y adopts halical-hairpin like pancreatic-polypeptide fold.;The structure propensity of predicted or experimentally determined protein structures as well as family of structures is examined via a comparative modeling approach. The evaluation tool developed within the framework of this thesis utilizes a novel complementary checking function, D 2Check, recently developed by our group. We have extended the D2Check analysis from the protein scale to that of the amino acids so to identify typical and atypical values of dihedral angles about a single residue (o--&psgr;) or those about two adjacent residues (psii-oi +1). At the residue level, a compact graphical representation is introduced to project dihedral angle compatibility of every amino acid (residual D2 score) of a given structure onto a color-coded strip. The color strip can be used to visually identify the typicality or atypicality of a given structure. This is possible since a particular structure is observed to be atypical only when most of its residues have atypical D 2 values (i.e. adopt unlikely dihedral angles). One can visually observe the likelihood of residue dihedral angles through a representation using color intensity to assess the propensities of the overall protein structure at a glance. The color strip difference strip, on the other hand, can be used to analyze structural similarities/ differences among protein families and structural effects of mutations. The color strip difference analysis of wild-type Staphylococcal nuclease (STN) for various LYS116 mutants has provided visual identification of the mutation site as well as other key sites that had been claimed as STN's biologically active regions. The D2Check methodology has been integrated into a web server (http://www.d2check.gatech.edu) to make the D2 code available to the scientific community. The server includes both protein level and residue level analysis and provides users with raw data as well as consequent graphs such as color strips, Ramachandran plots, position of the overall D2 score in the D2 distribution.;All-atom molecular dynamics (MD) has been extensively used to study motion of biomolecules (e.g. protein folding). However, conformational sampling of protein folding and unfolding events at the atomic scale requires substantial amount of computation and, this, is usually limited to shorter timescales compared to the real life events. Many methodologies, such as steered molecular dynamics (SMD), have been developed within the framework of molecular dynamics, to accelerate these events. SMD works by applying a series of time-dependent external forces on the system, for example on a model protein along a preselected unfolding pathway. When the system is driven through a path via external forces, it moves away from equilibrium. Jarzynski's inequality relates the applied force (i.e. work) to the potential of mean force (i.e. equilibrium free energy). For small systems with low energetic barriers, SMD in combination with the Jarzynski's nonequilibirum work relation yields accurate estimate of the potential of mean force (PMF) using a computationally accessible number of trajectories. For larger systems with higher energy barriers driven along extended paths, on the other hand, the applied force (thus required work) fluctuates dramatically across a very large range. Only the lowest energy trajectories dominate the PMF, and convergence would be achieved only through the determination of a prohibitively large number of trajectories. This can be surmounted by (i) increasing the sample size (as many as millions of realizations), (ii) decreasing the pulling velocity (as low as reversible velocity), or (iii) equilibrating the system at short intervals. All of these plausible solutions will, however, increase the amount of computation dramatically. This thesis presents a staged integration of the SMD methodology --- adaptive steered molecular dynamics--- that can be used to obtain a converged PMF efficiently. Each stage (or step) is designed to be short enough so that the work distribution exhibits good statistics and thus the corresponding PMF represents most of the generated trajectories. Adaptive SMD has been used to investigate the helix-coil transition of decaalanine in vacuum and in solvent and unfolding of several neurotransmitters (i.e. neuropeptide Y---NPY, peptide YY---PYY---and several PYY mutants) in solvent. The PMF along the stretching of decaalanine in vacuum was reproduced using adaptive SMD at much lower computational cost compared to conventional SMD. In solvent stretching of decaalanine using adaptive SMD has yielded an overall lowering of the PMF due to the stabilizing effect of the neighboring water molecules. The hydration effect is also confirmed analyzing the intra-peptide and peptide-water hydrogen bond counts. Adaptive SMD has also been used to calculate the PMF along the unfolding pathway of neuropeptide Y (NPY). Using this PMF and the activation energy barrier observed on it, the transition state rate of the unfolding of NPY has been calculated. The results show that monomeric NPY adopts the pancreatic-polypeptide fold as proposed by several experimental reports.