would they work for protein folding?

---

would they work for protein folding?

maybe you should explain what a evolutionary algorithm is...
since manny people use different words for the same things.

would they work for protein folding?

If you mean a Genetic Algorithm (which I think is a bit more of a specific term), then I also wondered that here, yesterday (a few posts down in the thread). As I said there, no doubt someone somewhere has tried using a GA to help with protein folding, and I'd love to know how it went!

---

Alver Valley Software Ltd - Contributing ALL our spare computing power to BOINC, 24x365.

would they work for protein folding?

If you mean a Genetic Algorithm (which I think is a bit more of a specific term), then I also wondered that here, yesterday (a few posts down in the thread). As I said there, no doubt someone somewhere has tried using a GA to help with protein folding, and I'd love to know how it went!

Yeah, that's what i meant, i think... sorry for missing that when you brought it up before.

---

Isn't the way how the folding works currently sort of evolving? The model doesn't try all possible structures as that would take too much time even with DC. But it checks different ways to continue from a start structures and works its way from there.

That's probably the main reason why why currently only do test runs for the model. If the model would test all structures then it would 100% find the lowest energy one finally. But since it doesn't test all but chooses based on some kind of evolutionary model we (well not exactly we) have to make sure that even though not all structures are tested, still the right one is found.

So I think such a model is already in use. But still somewhat in beta testing.

Warning: I'm no programmer, I'm just mouthing off.

From the name I assumed the code/algorithm itself would evolve.

Yes, I think that's what I had in mind. sorry for being vague, but i'm not really sure what I'm talking about to be honest, I just hoped it would spark discussion.

Something like the electric sheep screensaver, but that actually does work, and uses lowest energy as selection criteria rather then aesthetics. again, I don't really know much about this stuff, but I've heard it can be very effective for some things..

---

Okay, I thought it would mean the way to predict the correct structure would follow evolutionary strategies.

But I guess you're right, it often means the code itself evolves. But somwhow what we do currently is kind of that. There are many different models running on all the participants computers and only those models that come close to the real structure are taken into account while the others are discarded.

Something like a manual evolution.

I actually am a programmer, with a limited knowledge of GA. My understanding of GA is that it tends to have a lot more hype than real potential. I know that within the machine learning field, GA has its own set of devotees, and in very specific problems they can achieve good performance. However, as a generalized algorithm, or for protein folding in particular, I am not aware of there being any especially effective implementations.

One other thing to note is that GA's don't have "evolving code". There are a lot of misperceptions about GA and machine learning in general, and that's certainly one of them. It really only performs "feature recombination", and the underlying code base is entirely static (if the underlying code mutated as well, it would cease to function almost instantly).

Is there any reason for supposing that GAs could do objective function minimization more effectively for protein folding than other search algorithms? Your answer will depend on who you ask, but I personally lean towards no. Of course, it depends on the topology of the protein energy function, which I am completely unaquainted with, and I think most scientists are unaquainted with as well (since it's very complex).

---

thank you Mr. programmer.

so what if you apply one GA to another algorithm?
So you have a 'population' of separate, slightly different versions of Rosetta, run some predictions with them, select them based on lowest energy, then recombine them...

---

thank you Mr. programmer.

so what if you apply one GA to another algorithm?
So you have a 'population' of separate, slightly different versions of Rosetta, run some predictions with them, select them based on lowest energy, then recombine them...

There are a number of complications to such an approach. First, you'd have to define your "feature set" with an algorithm. That is, for any particular algorithm, you'd have to define what can be recombined with other algorithms. E.g. (in C/C++)

for(int i = 0; i < 10; i++)
{
// do computation
}

What can be varied in this expression? Do you make the "10" variable, do you make "computation" variable, do you make "for" variable, do you recombine machine instructions (quick path to disaster, though also the most generalized solution)?

You would have to have a very carefully designed scope for what can be modified and what can't.

That's a more technical concern, though. Apart from the amount of work it would take to do such a thing, you run into computation constraints very quickly. Rosetta, with only running 2-3 algorithms at a time, is already heavily resource constrained. It takes 80000+ models to get a somewhat decent minimal energy for a protein, given an algorithm. If you try to distribute the computation amongst 1000+ algorithms, you would significantly lose confidence in each algorithm's minimal energy. You would also have very low confidence that the lowest energy returned by a function is representative of that algorithm's capacity for generating low energy models. So I would expect (without having done any research on the subject) that any metric used to determine the "value" of a particular algorithm would have very high variance, and would therefore be unstable.

That's my guess anyway.

---

Looks like the new protein folding project POEM@Home makes use of GAs, see here:
http://boinc.fzk.de/poem/forum_thread.php?id=69#428

---

Looks like the new protein folding project POEM@Home makes use of GAs, see here:
http://boinc.fzk.de/poem/forum_thread.php?id=69#428

Certainly you can use some variants of GAs to solve protein folding. You will just stick to recombining parameters of the particular model, rather than the algorithm (which gets much more difficult). It's definitely possible to do this, I just said that I can't see any reason why it would be especially effective for this problem.

---

Looks like the new protein folding project POEM@Home makes use of GAs, see here:
http://boinc.fzk.de/poem/forum_thread.php?id=69#428

Certainly you can use some variants of GAs to solve protein folding. You will just stick to recombining parameters of the particular model, rather than the algorithm (which gets much more difficult). It's definitely possible to do this, I just said that I can't see any reason why it would be especially effective for this problem.

well it's worth a shot, innit?

---

Looks like the new protein folding project POEM@Home makes use of GAs, see here:
http://boinc.fzk.de/poem/forum_thread.php?id=69#428

Certainly you can use some variants of GAs to solve protein folding. You will just stick to recombining parameters of the particular model, rather than the algorithm (which gets much more difficult). It's definitely possible to do this, I just said that I can't see any reason why it would be especially effective for this problem.

well it's worth a shot, innit?

Sure, if someone's willing to throw a couple years at it, good for him/her.

---

Certainly you can use some variants of GAs to solve protein folding. You will just stick to recombining parameters of the particular model, rather than the algorithm (which gets much more difficult). It's definitely possible to do this, I just said that I can't see any reason why it would be especially effective for this problem.

This is kind of how my mutating genetic evolver works. I am stuck with further development without having a competitive low energy function.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.