Computer simulations of protein folding

This work explores possible approaches to performing protein folding simulations. Currently, due to computer resource limitation, only rudimentary models can be studied. While these simple models exhibit the behaviors commonly associated with the protein folding process, quantitative theoretical predictions for a specific protein cannot be made. We investigate the two major obstacles to the study of protein folding dynamics through computer simulations. First, it is well known that the folding time scale of a protein is inaccessible with traditional simulation algorithms. Experimentally observed folding time scales are reportedly on the order of a millisecond or longer, whereas the traditional simulation methods, such as molecular dynamics or Brownian dynamics, can only be used to simulate events whose time scales are on the order of a microsecond or shorter. To address this issue, we show, using a simple four-helix-bundle folding model, that the weighted ensemble Brownian (WEB) dynamics algorithm, a biased Brownian dynamics scheme, can be used to study folding events that occur on the time scale of one second or longer. The second problem is that the traditional molecular mechanics force fields often fail to distinguish between the native structure and deliberately misfolded decoys. As a result, these interaction potentials are not suitable for folding simulation and protein structure prediction. We investigate the empirical interaction potentials that are derived from the known protein structures. These novel interaction potentials are related to the distribution of inter-atomic distances in the known protein structures via a simple Boltzmann relation. Having derived these interaction potentials, we test them against a structure set that consists of current and superseded (outdated) protein structures in the Brookhaven protein data bank. These potentials prove very accurate in identifying the current, high-quality structures from their superseded counterparts, indicating that these potentials could be useful for protein structure predictions and folding simulations. While realistic folding simulations of proteins are not yet realized, we believe that the advances described in this thesis contribute significantly toward this goal.