Date: 18/01/2019 – 14:00
Candidate: Luciano Borges Censoni
Advisor: Prof. Leandro Martínez
Institute of Chemistry – Mini-Auditório
Protein chains exhibit a natural scale of discretization at the level of individual amino acid residues; no evolutionary mechanism, be it insertion, deletion, or local mutation, is able to affect the identity of a single individual atom in the chain, and the full diversity of protein structures and functions stems from combinations of the same standard amino acids in various chain lengths and orders. Here, we model protein structures as networks of interacting residues, and show that features of the topology of the resulting networks can be used to predict physicochemical properties such as the rate of folding and the ability to dissipate vibrational energy. In the first investigation, we employ modified molecular dynamics simulations to measure the intensity of the thermal coupling between each residue and the rest of the structure, and then show that node importance measures derived from network theory, particularly closeness centrality and eigenvector centrality, are well correlated to the observed coupling strengths. We construct an analytically solvable model of heat diffusion on a network, and show that the best estimates of the intensities of residue-protein thermal couplings can be derived from the Laplacian matrix that describes the interaction network. In the second investigation, we fit protein chains to a model based on self-avoiding random walks, and use it to de rive a probability distribution for the distance between amino acid residues as a function of their separation along the sequence. Using this distribution, we define an expression for the probabilistic information content associated to the relative position of each pair of residues in a protein structure. We then show that the average information content of all residue pairs in a structure is well correlated to the logarithm of its folding rate. Subsequently, we exploit the same measure of information content to identify redundant contacts, and show that we are able to predict a structure’s folding rate to good accuracy while taking into account less than 5% of its contacts. Finally, we implement a routine to calculate protein structural ensembles subject to geometric restraints derived from Nuclear Magnetic Resonance experiments, and show that the application of an optimization method based on the Low Order-Value Optimization strategy can help distinguish the restraints that correspond to the correct assignment of experimental resonances among decoys.