Hi,

As I understand it the current method is completely bruteforce, i.e. use the algorithm (and random numbers) to generate many structures.

I was thinking that you could use strucutres that have good pseudo-energies to try and generate more candidates, in a fairly typical optimization (global minimum search problem) sense.

For example, you could, say, have 10M structures generated and from those select the top 10 structures, and reissue them as part of the problem space (let's say that 1/2 of the people are folding the original protein at all times, to generate more good structures, and the other 1/2 are advancing already found structures, or at the beginnig 100% are generating the original, and then it drops to 90% and then 80% and so on). These top 10 structures are "unfolded" to the point where, say, only 10% of the folds have been made, and these would be the starting structures for the reissue. 10M structures later, and the top 10 are again selected, and taken back to the 20% location and reissued. (of course at the same time 10M more structures of the original protein are completed, and the top 10 of those enter the 10% phase).

It would basically be a pipeline:

0%-->10% --> 20% --> 30% --> ... --> 100%

In terms of viewing it as a global minimum problem, this should (statistically!) help out quite a bit. After all, a good structure probably is about average for the possible folds along its path. This method basically envolves refining each stage to produce a more optimal solution.

Logistically, it would probably be a bit more complicated to run things, but it shouldn't be too hard.