Home » Technology » Scientific Computing » Simulation Studies of Protein Folding

Simulation Studies of Protein Folding

© Adam Hughes

One of the holy grails for simulation scientists in the chemistry and biology fields is gaining a detailed understanding of the mechanisms involved in the folding of a protein molecule. This phenomenon is commonly known to occur as part of defining the functional work a protein does in a biological system, but precisely how it happens has remained something of an unsolved mystery. A robust computational approach should be able to provide some explanations, but at several hundred atoms for the protein itself and several thousand more for any solvent that may need to be studied, the cost has been prohibitive. In addition, the time scale on which this happens is quite large in computational terms. To overcome these limitations, it has been necessary for researchers to employ the absolute state-of-the-art in both algorithms and computing resources.

A protein consists of a backbone of amino acid building blocks, plus any number of modifying chemical groups at various places along this spine. When a new protein is formed, it is in its extended state, with its constituent peptides in a more-or-less straight line. After formation, the protein begins to fold, its final configuration depending on the amino acids of which it is built, as well as the nature of its surroundings, such as the pH of a solvent. The goal of studies in this field is to understand exactly what determines the folded structure of a protein, which in turn dictates its function in life.

Obviously, it would be wonderful to be able to conduct very detailed quantum chemistry studies on these species, but the thousands of atoms in such a system quickly rules out that possibility. Even traditional molecular dynamics (MD) methods, which are designed to handle systems of large molecules or with large numbers of smaller molecules, easily fails at this task because of computational limitations. Protein folding is generally believed to occur on a scale of milliseconds to seconds, and traditional MD is usually good for a nanosecond or so of simulation in a reasonable amount of time.

Obviously, new methods have been required to approach problems of this magnitude. There have been attempts to force the "virtual protein" (protein model) to complete its folding much faster than it does in reality, thus providing some insight into the problem in a more tenable period of time. Other algorithmic advances have been employed in approaching this problem, too, but probably the most significant advance has been the full exploitation of our nation's massively-parallel computers. Utilizing several to several hundred processors to perform a calculation can provide a very prominent speed-up, and make simulation applicable

The copyright of the article Simulation Studies of Protein Folding in Scientific Computing is owned by . Permission to republish Simulation Studies of Protein Folding in print or online must be granted by the author in writing.