science man wrote:Hey I have my own question that goes along with this topic and that is, do we have to study each cell at a time?
I'm not sure I understand your question. Folding@home isn't in the business of studying cells, but the folding of proteins, which are (some of the) constituents of cells (the things that do the work and provide the structure).
Is that why the ultimate project is taking so long even with all the cpu's on the network? If so, why? There is only of blood cells we need to study to do this. Why can't we just study one cell and logically assume its results apply to all the other cells in that type?
The reason this "is taking so long" is that protein folding is amazingly complicated. There is an picture due to Levinthal that illustrates this very well. As you may know, proteins are long chains of amino acids, each of which may adopt numerous different conformations. For simplicity, let's assume each amino acid can adopt two distinct conformations (this is a lower bound on the true number -- it's a very coarse representation of a somewhat complicated object). Then if there is a chain of 100 amino acids, the protein chain itself can take on 2-to-the-100th-power distinct conformations. That's
1,267,650,600,228,229,401,496,703,205,376
different conformations (oft referred to as "a buttload"). Now, we don't search through all of those on Folding@home, but we certainly search among them not only for those of the 2-to-the-100th-power conformations that correspond to the folded or "native" conformations. And it's further complicated because we're interested in the
pathways the protein chain takes (what conformations it visits) on the way to being folded.
That takes a very long time, especially since (1) a protein moving from conformation to conformation according to the laws of physics can take a long time (in terms of calculations) and (2) it's necessary from a physics standpoint to get good statistics on all this, so similar calculations have to be repeated multiple times.
And that's just for one protein of 100 amino acids. There are something like (someone please correct me if I'm mistaken) 100,000 distinct proteins in the human proteome. Now, we're not trying to fold all of them: rather, there are some obvious medical targets that we'd like to understand (like the a beta peptide, implicated in Alzheimer's disease) or ones which are experimental models that we'd like to understand (like the villin headpiece).
A further complication is achieved in that there are a lot of approximations made in our calculations (which are generally required to make the models tractable). These approximations introduce unknown statistical errors into the calculations, so we're forced to check the results not only against experiments, but with slightly different models.
What this all comes down to is, we hope that with enough calculations, even with all the complexities and approximations, that we can come up with *general principles* of protein folding so we don't have to calculate them all. I'm not likely to figure this out myself this afternoon, let alone in the next two years, but back to work anyway!
Dan
.
