Slippery slope argument

[dbuske]

I was thinking that from watching the folding that it tends to go down the slippery slope over time. You might get a 12.xx RMSD and then never get anything lower. It seems to just get worse over time.

Just throwing out an idea here: Maybe if someone gets a lower RMSD, that fold could be sent to all folders and used as a basis for further folding. Then possible someone might improve upon it by chance. Those that get worse then start on that same fold after lets say 10 generations of worse folds. They just repeat until someone gets a better fold and they are sent a new lower RMSD.

[Galuvian]

That would be one way to massively distribute a genetic algorithm. But unfortunately protein folding is not that clearcut. The peaks and valleys are intentional. Imagine that you are in the middle of a desert, with sand dunes all around. The only way you can find out if there is a lower valley over the next dune is to climb the dune and look.

[dbuske]

They have tried the "wanderers in the desert" path searching up and down the dunes aimlessly for two years, I am told. Using your desert wanderers analogy:
Maybe they should try giving each wanderer a walkie-talkie and when one sees an oasis, they check their compass and radio the others the coordinates. Thus everyone heads initially in the correct direction.
Some will again head off in the wrong direction in a little while, but someone would go on to find the correct path and be closer to getting out of the desert. Then he or she radios in that coordinates.

[Pascal]

No, 'tis rather this:
No one of them has a walkie-talkie, many of them find an oasis. 'Tis the question, which of these oases is the best one. It's better, if everyone searches for himself, some have a good map (many/fast computers), others a worse one (single, slow machine).
Randomly, also a wanderer with a bad map has a chance to find the right oasis.

[dbuske]

The desert wanderer analogy does take into account the different paths to finding the right shape of the protein. They are just getting a little push to a better path now and then. I think it would be better this way than wasting alot of computer time on lost wandering.
When the mainframe would download the new best RMSD, then the computers would start folding again and diverge over time, but they all started at the same point.

[Paratima]

Before you go too much further, it might be helpful to read the previous discussions on this very subject in the Educational section (http://www.free-dc.org/forum/forumdisplay.php?s=&forumid=32) .

[Brian the Fist]

Yes, the phase III genetic algorithm (which presently exists only in my head) will work something like this and is discussed elsewhere.

[AMD_is_logical]

Originally posted by dbuske
Just throwing out an idea here: Maybe if someone gets a lower RMSD, that fold could be sent to all folders and used as a basis for further folding. ... You can only calculate the RMSD if you know the "answer" (the correct fold). We are currently working on proteins where the answer is known because we are trying to get the algorithm working. The algorithm will only be useful if it can find the correct fold of a protein where the answer is NOT known in advance.

Thus, the algorithm must work when the answer isn't known in advance, meaning we can't make use of RMSD to fold the protein. We can only use RMSD to evaluate how well the algorithm does with proteins whose correct fold is known.

[dbuske]

I see what you mean. It would in effect be using the actual RMSD to guide the folding and no longer be a blind test.

[dbuske]

Well I don't think the idea is dead yet. They could possibly use the quick energy test instead. I wonder what Howard has in mind?