Theoretical and computational studies in protein biophysics: Folding, information, and networks

For the prediction of protein structure, we present a simplified form of threading, Using extreme value statistics, we show that the trade-off between the search space size and the optimal alignment accuracy can be captured in a single statistical parameter epsilon. The parameter epsilon enables us to confirm that the best threading results can be obtained by a combination of just two or three gaps.; For coarse-grained models of protein folding, we present a designed all-atom potential of mean force for estimating free energies in protein folding. The novel atomic potentials perform well in a Z-score and fold recognition test. Comparison of different protein conformations under the various atomic potentials reveals a high correspondence in the contact types that make the dominant contributions to the estimated free energies. This consistency may be interpreted as a sign that the design procedure is extracting physically meaningful quantities.; For the folding of proteins with a frustrated, realistic potential, we devised a residue-specific, protein backbone move set for efficient sampling of conformations in folding simulations. We show the knowledge-based move set enables more efficient finding of ground states, and faster crossing of free energy barriers when compared to the standard move set. Use of this move set for calculating thermodynamic quantities is discussed.; For investigating the information basis of protein to protein recognition in cellular networks, we measure the information content at 895 protein interfaces. A simple phenomenological theory reveals a scaling connection between interface information and the free energy of association. Our theory also suggests that mutual information in contacts emerges by a selection mechanism. We verify this by showing a statistically significant correlation between conservation of the twenty amino acids and their individual contribution to information content.; For a description of noisy gene network dynamics, we present a Langevin model. We solve this model exactly using a path integral technique and derive several predictions. We propose experiments by which these predictions might be tested. Microarray data taken from Saccharomyces cerevisiae show correlations in expression fluctuations have a highly statistically significant dependence on gene function, and exhibit a remarkable scale-free network structure.