Multiscale Protein Networks: Interactions, Two Component Systems and Kinetics

Living organisms are driven by molecular machines called proteins. To understand the molecular mechanisms of the cell, it is not enough to understand individual properties and functions of proteins but rather uncover the interactions between these molecules. This work presents a multiscale approach to study protein networks. The top scale introduces an algorithm for protein and domain physical interaction estimation. This algorithm is validated in the model organism S. cerevisiae and applied to the soil bacterium M. xanthus. Interactions are predicted in the reversal mechanism of this bacterium, proposing an extended model for reversal control and motility engine activation/deactivation. At a second scale, this work investigates the problem of specificity residue identification in a bacterial signaling pathway called Two Component System. Using a method of subgraph identification and learning, specificity residues or 'graphlets' are proposed to estimate factors contributing to domain-domain specificity in histidine kinases and response regulators. This network based approach reproduced experimental studies in E. coli and produced a series of hypotheses of domain specificity that could extend our knowledge of interaction interfaces in two-component systems. At a finer scale of individual atomic protein structures, this work introduces a methodology to identify metastable clusters of atomic conformations obtained by combining and correlating Nuclear Magnetic Resonance (NMR) experimental data with molecular dynamics simulation. This methodology provides a way to find a network of metastable states (nodes) and their most visited transitions (links). This represents a mechanistic view of conformation change in a given protein. This analysis is applied to the WW domain of human protein Pin1. In addition to high correlation with NMR exchange rates, the network of kinetic macrostates identified here proposes that Pin1-WW domain in unbound state visits its bound configuration through a state that corresponds to a minor population of relaxation dispersion experiments. This computational approach provides a detailed structural view of conformational change that NRM experiments are not able to identify. This dissertation introduces novel methodologies that were validated using experimental data and can be generally applied to different organisms. These methodologies propose testable hypotheses that could improve our knowledge of biological problems and systems.