|
|
SUMMARIES OF SELECTED SCIENTIFIC JOURNAL ARTICLESNature 2002 | JMB 2002 | Physical Review Letters 2001 (This is a simplified summary of the article by Michael R. Shirts and Vijay S. Pande in PHYSICAL REVIEW LETTERS - 28 MAY 2001 titled MATHEMATICAL ANALYSIS OF COUPLED PARALLEL SIMULATIONS ) Proteins are initially in an unfolded, relaxed state like an expanded spaghetti noodle. Due to several forces in action, in the cells they begin to fold very rapidly into a 3 dimensional structure to serve their biological function such as carrying oxygen from lungs, forming the structure of hair. The structure of initial unfolded state and the final folded stage can be determined through various well established methods. (If interested in the explanation of the method follow this link x-ray diffraction) We would like to understand the folding process itself which will have many benefits for all of us as explained elsewhere in this site.
Pande group Computer Simulation The fastest proteins fold in about 10 microseconds (1 microsecond is one millionth of a second). Even the fastest computers could only simulate few nanoseconds of this process (a nanosecond is a billionth of a second). The reason for this is there are a large number of states proteins can go through until they settle in the final folded state. If you take a typical coin toss, there is only two possibilities: heads or tails. In a dice roll there are six possibilities. In the case of proteins there are so many possibilities since there are 1000's of individual atoms, and even in a simplified simulation there are at least 5 different type of forces acting upon each atom. Even worse proteins can fall into "traps" which may take a long time to get out. A single computer is like a single lane road, no matter how well the road is made in heavy traffic the cars will not be able to move very fast.
If you have a road composed of many lanes the traffic will move much faster. This is the idea behind parallel simulations. When you participate, your computer becomes one of the lanes in parallel simulations. The more computers participate at Folding@Home, the more lanes of simulations are available to us.
Even though a sample protein may take a long time to fold to simulate by a single computer, if we run 10,000 or more parallel simulations independent of each other we may expect to observe a small number of proteins to fold. For example, if we run 10,000 simulations simultaneously we could expect about 25 about those simulations showing folding. After the simulation we could get very valuable information, data from those that folded during our simulation.
APPENDIX - Mathematical description of parallel simulations In a typical protein folding process, if we plot percent folded over a period of time we get a graph like below (%folded = 1 - exp -t/T where t is elapsed time and T is the average folding time):
Fastest proteins fold in10 µs (1 µs = 10 x 10 -6seconds or one millionth of a second). A single computer can simulate a nanosecond in a day(one nanosecond = 1 ns = 10 x 10 -9seconds or billionth of a second) because of all the different possibilities exist for an unfolded protein until it reaches the final stable, folded state. So using a single computer it will take about 30 years to simulate even the fastest folding protein. The good news is 10 µs is the total time, and within this time span some proteins may fold much faster than others just by chance. If you examine the graph below, the initial percentage of folded proteins can be approximated with a linear relation (%folded = t/T)
In the first 10 nanoseconds (one nanosecond is a billionth of a second) of simulation, ratio of folded proteins to the total number of proteins approximately equal to elapsed time divided by the total time it takes for all the proteins to fold. So in the first 10 ns, if we approximate %folded proteins to be equal to t/T and if t = 10 ns and T = 10 µs then: % folded = t/T = 10 ns / 10 µs = .001 x 100 = 0 .1% or in other words in the first 10 nanoseconds about one thousandth of protein sample may have folded. If we run 10,000 parallel simulations then: 10000 x .001 = 10 folded proteins. So with 10,000 parallel, independent simulations we can approximately simulate about 10 folded proteins. Alternatively, if we know the percent of proteins folded, and the time it took, then using above equations we can determine the folding time constant T for the protein to be compared with experimental results. This is one way to validate the computer simulation approach experimentally. As the graph below shows, there is a close agreement with folding time constants determined by computer simulation and experimentally.
|
