Page 1 of 1

Supercomputer sets protein-folding record

Posted: Thu Oct 21, 2010 12:11 am
by deregtx
http://pubs.acs.org/cen/news/88/i42/8842notw1.html

How does it compare to F@H ?
I hope you guys are working together? Because it would be a gigantic waste of time and resources if that supercomputer is redoing all the work we did.

Btw they are donating these supercomputers to universities so maby you could ask for one? :mrgreen:

Re: Supercomputer sets protein-folding record

Posted: Thu Oct 21, 2010 1:45 am
by VijayPande
In many ways, I think they are still several years behind us. If you take a look at the proteins they simulated, it's the systems we first folded several years ago (eg villin). The systems we're publishing now fold 1000x slower than villin and they would not be able to study the folding of those systems with their current approach. With that said, it's really neat to see what dedicated hardware can do and they've done very nice work.

PS I think it's wrong to say that this supercomputer has "set a protein folding record." I also don't think they claim that.

Re: Supercomputer sets protein-folding record

Posted: Thu Oct 21, 2010 2:00 pm
by Qinsp
Outside observation (ie - clueless layman):

The distributed computing model has an advantage over dedicated hardware for long term projects. The DC "hardware" is self-improving. Supercomputers take years to design and fabricate. Often their core technology is dated somewhat at the release date, and doesn't get better with age. If my reading is right, the fastest processors in Anton are 800mhz. If they started designing it today? Maybe 1200mhz would be used?

I've been using computers since 1975. What's on the shelf in Walmart today is thousands of times more powerful than high end desktops of the late 70's. 64k of RAM vs. 4,000,000k of RAM, ~2mhz? single core no math-coprocessor vs. 3,000mhz times 4 cores. It really wasn't that long ago that a computer at home was a novelty; a gadget to goof off with. Nobody could come up with a real reason to have one other than to have one: "Uh... I uh... use it to balance my checkbook! And write letters! Sure, my Apple II cost $3500 but it's a real money saver..."

And who would ever have thought that the silly Pong game console would eventually be a "supercomputer" in it own right? Could you imagine firing up a PS3 with Call of Duty on it in a room full of people in 1980? They would have thought it was magic or a parlor trick. All the computers in the world combined could not have done real-time 3D motion of multiple complex objects. They probably couldn't even do the sound in real-time.

I just took my Intel 80486DX-33 out of service last month. I bought it when they were first released (~1990?), and ran it 7 days a week for 20 years. At the time, CAD was perhaps the only reason to have one. They were so fast that you had to slow them down to run many programs. My cellphone today is hundreds of times faster than that "monster" 486 was.

What will be on our desktops in 2020? 2040? Anybody's guess. Graphics processing seems to be growing faster than CPU technology is growing right now. Perhaps because there aren't any normal apps that are demanding faster CPU's? One thing that is a safe bet, is that a supercomputer made today will not compare with one made in 2020.

Professor Pande and others have proven the DC model is viable and stunningly powerful. And it will get better with age.

Re: Supercomputer sets protein-folding record

Posted: Fri Oct 22, 2010 7:35 pm
by deregtx
VijayPande wrote:In many ways, I think they are still several years behind us. If you take a look at the proteins they simulated, it's the systems we first folded several years ago (eg villin). The systems we're publishing now fold 1000x slower than villin and they would not be able to study the folding of those systems with their current approach. With that said, it's really neat to see what dedicated hardware can do and they've done very nice work.

PS I think it's wrong to say that this supercomputer has "set a protein folding record." I also don't think they claim that.
Well the news says its a new record but i dont know :mrgreen:

I just would like to see them work together with F@H because i dont like that so many projects trying to do exact same thing...

Re: Supercomputer sets protein-folding record

Posted: Fri Oct 22, 2010 10:27 pm
by OzPapaSmurf

Re: Supercomputer sets protein-folding record

Posted: Fri Oct 22, 2010 10:44 pm
by bruce
The computer performed two types of simulations of well-studied systems: In the first, Anton's 100-microsecond simulations of the protein domain FiP35 show that the domain folds and unfolds in the same progression of steps each time—a surprise to some researchers.
I wonder just how many Runs/Clones they ran to verify the statistical validity of the claim that this is always true, or just worked for the particular case(s) that they ran.
The second simulation involved a longer, one-millisecond study of the dynamics of the already folded protein bovine pancreatic trypsin inhibitor (BPTI), which explored the intricate motions of protein in water. Results were published in Science (2010, 330, 341).
See also http://folding.stanford.edu/English/Papers#ntoc4
Anton was built solely for the purpose of molecular dynamics (MD) simulations of the behavior of large biological molecules. The machine, which was completed more than a year ago, stakes its claim to fame on its ability to perform lengthy simulations that aren't possible on current supercomputers. The new results mark the first significant test of Anton's prowess.
Note that he carefully compares their system to other supercomputers, thereby excluding virtual supercomputers like FAH. That doesn't minimize what he has actually achieved but I still don't see that we have a scietific "record" for MD. Where did that work come from? Maybe I should remove it from the title.
MD simulations, which track the motions of every atom in a large molecule, are limited in their time scales by computer power and architecture. It can take months for a supercomputing system to simulate only tens of microseconds of a protein's dynamics. But many proteins fold on the millisecond time scale.
That's an extremely important statement. We have to keep reminding the FAH donors of what that means from the perspective of a Donor.
Anton, its creators report, is capable of performing millisecond-long simulations of systems containing tens of thousands of atoms in 100 days—100 times faster than with current supercomputers.
Again comparing to current supercomputers.

I suspect that he might be using GROMACS but that's not stated. He does seems to be using explicit solvents.
Once a simulation starts, it can't be adjusted. "You cannot change conditions on the fly; it has to be hands off," Schulten says. Only about 10% of protein-folding problems can be done this way, he adds.
So it's hard for him to generate good statistics ... one Run/Clone at at time. If a protein folds correctly 99% of the time, FAH has to analyze enough different possible folding sequences to have some of them show up in the 1% ways that it might actually mis-fold.
Recently, Shaw Research donated another Anton machine to the Pittsburgh Supercomputing Center for research use by universities and other nonprofit institutions through the National Resource for Biomedical Supercomputing.

Schulten's group is one of 45 that were allotted time on the machine by the National Science Foundation. His group began simulations last week, and "it's working very well," he says.
Collaborating with FAH might be a benefit to both projects. In fact, all 45 might do well to collaborate. Presumably some of that happens at the scientific conferences that they attend.

I'll bet some of the PG researchers would be happy to get their hands on some supercomputer time to augment what the donors contribute though FAH. I'm sure that the NSF considers many researchers, though, and might choose to allocate their money to somebody who DOESN'T have access to FAH.

Re: Supercomputer sets protein-folding record

Posted: Sat Oct 23, 2010 12:44 am
by Hyperlife
bruce wrote:That doesn't minimize what he has actually achieved but I still don't see that we have a scietific "record" for MD. Where did that work come from? Maybe I should remove it from the title.
This article, which was linked from Slashdot, claims that the millisecond time scale was "more than a hundred-fold greater than the previous record." I thought F@H already did that almost a year ago.

Re: Supercomputer sets protein-folding record

Posted: Fri Aug 12, 2011 2:24 pm
by Jorge1950
The WU processing ultra-large to Anton; It is undervalued.
I loved processing units Advanced. I was a mathematician and I liked the theme of the new methods.
But my productivity dropped to 40%, 100% original.

Be subvaloran in the extreme points of Advanced units. Perhaps it is the right policy, I do not know these details.

I retired from the Advanced WU to the CPU; and I would advise to withdraw. I am studying the case of WU to GPU, Advanced.

Re: Supercomputer sets protein-folding record

Posted: Fri Aug 12, 2011 3:17 pm
by MtM
Jorge1950 wrote:The WU processing ultra-large to Anton; It is undervalued.
I loved processing units Advanced. I was a mathematician and I liked the theme of the new methods.
But my productivity dropped to 40%, 100% original.

Be subvaloran in the extreme points of Advanced units. Perhaps it is the right policy, I do not know these details.

I retired from the Advanced WU to the CPU; and I would advise to withdraw. I am studying the case of WU to GPU, Advanced.
I don't really understand your post.

Unit's advanced -> -advmethods or high performance clients? Theme?

Subvaloran means undervalued, but I don't understand this sentence at all -> Be subvaloran in the extreme points of Advanced units. Perhaps it is the right policy, I do not know these details.

Advanced WU to the cpu -> smp/high performance client or -advmethods? What are you advising and why?

Re: Supercomputer sets protein-folding record

Posted: Fri Aug 12, 2011 4:39 pm
by patonb
He used a translator... I think hes just found out about the lower bigadvanced pts and is switching to regular smp.

Re: Supercomputer sets protein-folding record

Posted: Wed Oct 12, 2011 9:07 pm
by Jesse_V
Just started work on project 7610. http://fah-web.stanford.edu/cgi-bin/fah ... ned?p=7610 Certainly a very interesting project description, as it explains that F@h is testing itself against Anton. I'm looking forward to the results, but it seems to me that F@h's Markov State Models will turn out to be right on the mark yet again. :)

EDIT in 2012: See Dr. Pande's blog post here and I've strived to summarize the F@h-Anton relationship here as well, including some of the points covered in the publication Dr. Pande linked to below.

Re: Supercomputer sets protein-folding record

Posted: Wed Oct 12, 2011 10:50 pm
by VijayPande
PS Here's an update of our work in this area
http://pubs.acs.org/doi/abs/10.1021/ja2 ... istoryKey=

I'll write a blog post with more details on this in a few days.