When I ask foldit to mutate my protein, how does foldit determine the structure? My guess is that if I select 4 residues, and mutate them to any amino acid type, then foldit will not compute the energy of each of the 202020*20 = 20^4 structures, and pick the best. I think there is some special search strategy so that it takes less time to find the best scored structure.
How can I find out?
I'm curious about this, too.
hi gwoolly,
In answer to your question, we use a different search strategy than complete enumeration of the side chain identities. You are correct that it is very computationally demanding to find the best mutations using complete enumeration. To circumvent this cost, we use something called a 'monte carlo' search strategy to find the best combination of mutations (see wikipedia "monte carlo method" for an overview). The disadvantage is that we are never sure that the mutations chosen are the absolute 'best'.
The original paper describing the monte carlo search strategy for the Fold.it system is described in:
Brian Kuhlman and David Baker, PNAS (2000) vol.97 page 10383.
best of luck,
tim
is this means if we do mutate few times form same start point we can expect different results (using do_mutate on few segments at once)?
Do shake also use monte carlo?
We found that wiggle is repeatable (deterministic, not random):)
My guess is that shake also uses the Monte Carlo method– there are multiple instances where I could manually double-click on a side chain after shaking and still gain points. As for wiggle, my guess is that it's repeatable because it is technically "gradient-based minimization" according to the Nature article they published last year…