Mike H

It should probably be mentioned that being able to predict the secondary structure from the protein's primary structure is somewhat better with regard to accuracy, although as usual this also tends to improve if you're working with homologous proteins for which you have experimentally determined structures. After all, you pretty much are limited to predicting whether or not a sequence adopts an alpha helix, beta strand, or loop. Branden and Tooze (Introduction to Protein Structure) has a blurb on this in their chapter on prediction and engineering of protein structures if anyone wants to get the abbreviated version.

Probably one of the more interesting and useful ideas out there is trying to utilize minimal data sets (usually from NMR, although EPR would work as well) in defining constraints for molecular modeling. While determining structures explicitly by NMR (solution or solid state) is a non-trivial task, especially as the size of the protein or multisubunit complex increases, it's a bit easier to extract out some distance constraints and employ them in the modeling work.

Godbert

yes, that makes sense. Could also be used to confirm that your calculated structure (once we are capable of getting one) is indeed accurate.

lpetrich

It is indeed possible to find a protein's "preferred" folding with the help of a computerized version of a ball-and-stick model, but for a typical-sized protein, the amount of computation necessary is enormous. Which is the great difficulty.

In fact, the folding@home project does exactly that, farming out the computational labor to participants' computers in the fashion of seti@home.

Godbert

Just to make sth clear: it might be possible (although I doubt the software is ready yet) , but as far as I know it has not been done yet - at least not for any protein containing a new fold.