Using molecular dynamics for high-resolution protein structure prediction

We now live in the genomics era, where novel sequences abound, awaiting structural determination that will probably only ever be solved experimentally in a small fraction of these new targets, due to the time constraints of experimental methods. Thus, the allure of accurate, insightful protein structure prediction is greater now than it ever has been, but the leading-edge methods fall well short of providing useful predictions, unless there is a very high percentage of sequence identity. Amino acid sequences exhibit an enormously large number of possible conformations, leading to a dimensionality problem that can only be overcome by reducing the representation of the protein. Unfortunately, resolving the difficulty of dimensionality by simplifying the representation also limits the extent of accuracy that can be had. A logical answer to this predicament is to pass on structures obtained from an ab initio or comparative modeling protein structure prediction effort, which both are effective at dramatically reducing the number of allowable configurations, into a more accurate method such as molecular mechanics/dynamics, to move from low/medium resolution structure predictions to high resolution ones. This can be accomplished by simply more effective scoring of the large number of predictions that arise from the early stages, and by drawing the best predictions ever more closely to the native state. This thesis has been an exploratory effort, met with significant success, designed to evaluate the promise of using methods within molecular mechanics, molecular dynamics in particular, in the endgame of protein structure prediction for high resolution protein structure prediction.