Coarse grained Protein folding in under 10 minutes

[b1llyb0y]

Just came across this, would leave to know If / how this effects FAH?

http://www.sciencedaily.com/releases/20 ... ce+News%29

Apologies if this has already been posted ..

Edit by Mod: Two topics merged

[John_Weatherman]

http://csb.cs.mcgill.ca/tfolder/ So that's it then folks. You can turn off your machines and Vijay can just do all the work on the webpage! That's progress for ya!

[MtM]

From that last link..

tFolder is a program that enables you to compute coarse grained representations of the energy landscape of β-sheet proteins and to predict their folding pathways. All you need is to enter your sequence and select the maximal number of strands allowed.

So.. is FAH aimed at Beta-sheet proteins -> http://en.wikipedia.org/wiki/Beta_sheet

The higher-level association of β sheets has been implicated in formation of the protein aggregates and fibrils observed in many human diseases, notably the amyloidoses such as Alzheimer's disease.

Seems so.

So, OP questions seems very legit to me, and surely deserves more then the jesting response given above.

While it probably won't replace this project, if the science is right it's not unthinkable that maybe a new core can be developed following the fundamentals described in the paper 'Efficient traversal of protein folding pathways using ensemble models'.

But, if that's feasible and if it would complement the already existing core's that's not for me to say, I have no clue

[toTOW]

The program is called tFolder and I found an article that claims it might mark the beginning of the end of the need of distributed computing : http://www.techno-science.net/?onglet=news&news=9237 (it's in French, so you might need an translator ).

Could FAH take advantage of this ? Is it really the end of the distributed computing needs ?

Unfortunately, the source is not very well documented, so maybe someone else already heard of this and could answer my questions ...

[tjlane]

b1llyb0y, thanks for the article! I just DL'd the primary citation, and it makes for an interesting read.

The algorithm reported here - tFolder - is interesting, but much in the same vein as "coarse-grained" work that has been around for a while. It is very fast computationally, but also very limited in scope: it can only predict B-sheet proteins, and a very coarse description of what might be going on in protein folding. The smallest relevant units for protein folding are atoms, which means that the most general physics-based models that will be completely general should have atomistic detail. The all-atom simulations that FAH makes possible are much more general (we can do stuff like look at ABeta aggregation, which coarse models often can't do), compare quantitatively to experiment, and really get at the physics of *why* proteins fold, something coarse-grained models can't really do. Unfortunately, all-atom simulations are very expensive, which is why FAH is so critical to getting a bunch of people with a bunch of computer cycles all working together on these problems!

[kasson]

In a word, no. Such coarse-grained representations are typically not very accurate at predicting the kinds of questions we'd like to answer. We haven't evaluated this particular claim, but such methods tend not to be suitable as an adjunct for FAH let alone a replacement. We have used FAH for some more detailed coarse-grained simulations (e.g. Kasson et al., PNAS 2006, Kasson and Pande PloS Comp. Biol. 2007)--coarse-grained simulations tend to be much faster than atomically-detailed ones, but there is no "free lunch." In our evaluation, the more coarse a representation is, the fewer questions it is usually helpful in answering. (That said, a coarse-grained representation can certainly be useful and in fact the most appropriate way to answer *certain* questions. It's just not generically good.)

[toTOW]

Thanks for the answer

[b1llyb0y]

Thanks for the input, I thought I might put it out there and see what the FAH team thought .. could this be used by FAH in the future first to predict a rough fold and then as per usual fine grained folding?

If I remember correctly, the author(s) are trying to branch out in to other proteins ..

[tjlane]

b1llyb0y -

That is actually an interesting idea. In order to do so, the method would have to develop quite a bit further. We've actually taken to doing something similar using Rosetta, the Baker lab's structure prediction software, as a method for getting an idea of what different conformations a protein can adopt, then starting our simulations from all those structures. We're hopeful that tactics like that will allow us to fold more efficiently, and reach for bigger and bigger targets (= more biologically relevant)! My new project 7600 (coming to full FAH soon) is one of the first to use this method, so we'll see how well it turns out!

[bruce]

I would think that one of the problems with this approach is that every conformation has a certain degree of stability. That means it takes a certain amount of time before it leaves that conformation and moves to another (except maybe for the final conformation with the lowest energy, depending on other factors). The less-stable conformations might never be discovered with coarse-grained folding, and perhaps misfolding only happens if the protein passes through a particular less-stable conformation.

Exploring all of the possible conformations takes lots of computer time, of course, and if you get enough information from Rosetta to do it faster and still cover all relevant regions of the conformation universe, then GOOD.

[tjlane]

Bruce is absolutely right that the big advantage of FAH-style simulations is that they give full atomistic detail (think as close as you'd ever want to zoom in with a perfect microscope) in a very general way. Doing atomistic simulations that capture kinetics (ie not just protein conformations, but also interconversion between those conformations), we know we're not missing anything. All of that comes at computational expense, though, which is why FAH is necessary.

Rosetta only gives us *structures*, and not *kinetics*, so it is a starting point, but definitely not the entire picture.