"Is Distributed Computing being Distributed Badly?"

Message boards : Cafe Rosetta : "Is Distributed Computing being Distributed Badly?"

To post messages, you must log in.

Previous · 1 · 2

Profile nasher

Send message
Joined: 5 Nov 05
Posts: 98
Credit: 618,288
RAC: 0
Message 19682 - Posted: 2 Jul 2006, 11:18:14 UTC - in response to Message 19582.  

yes there are lots of problems with the article on slashdot

and most the responses are silly..

personaly i do read slashdot almost daily but i normaly only use it to read the main articles and dont get caught up with the junk posts that are added.

as my own opinion.
is Seti@home a waist of CPU time. yes i think it is BUT its your computer
and its your computer time use if how you want. its a setpping stone
for alot of people to get into distributed computing.
Are the BIG PHARM COMPANIES gona get rich from our work here and such
Probaly, once we here manage to get all kinks worked out and manage
to find cures for diseases then they will JUMP on the bandwagon and
start using the reserch to make cures. but thats kinda the point
isnt it... to find cures. its just also as a byproduct gona make
some people rich also(wish i was one of the rich ones)
When people ask me what project I think they should run on there computers this is what i tell them.
To me it dosnt matter what program you run as long as you are running one you enjoy for some reason or another. Yes there are some projects that have more scientific results and some that are silly as well as some that in my opinion wont produce anything worthwile but again its your computer and your cpu cycles you are donating. as long as you enjoy the projects then its the right project for you.

Personaly I have made it a gole of mine to try and run a little of every project and then decide on my own. I love Rosetta and I used to love the climate prediction (projects are WAY TO LONG) but every project has its own little set of users that to them its important and I respect them for at least running a DC project. Over the years i have tried almost everything out there and curently Rosetta is my main focus.. but i keep serching to see if something i am more intrested in shows up. No clue what it is but i will know when i see it i hope

till then Happy DC ing
ID: 19682 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Dimitris Hatzopoulos

Send message
Joined: 5 Jan 06
Posts: 336
Credit: 80,939
RAC: 0
Message 19712 - Posted: 2 Jul 2006, 20:53:52 UTC - in response to Message 19467.  
Last modified: 2 Jul 2006, 21:42:26 UTC

Tarx, I found some time over the weekend to go through most of the slashdot conversations (waste of time you will say, but it satisfied my curiosity :-)

You sent some comments from slashdot:
"I'm a guy who was once associated with one of labs/projects mentioned above. I was working on the problem for years, and have a great deal of expertise in the area. I can also tell you that the project is complete and utter crap, from a scientific perspective. The PI routinely misrepresents the project goals, claiming "possible" results that could never, ever come from the type of research performed. In general, the "science" is poorly-conceived and improperly controlled, and most of the "experiments" are methodologically flawed. I can't post my name here...it would be career suicide. As one of the authorities to whom you seem so desperate to appeal, let me assure you: if you are devoting your resources to this project, the world would be a better place if you simply turned your computer off."
Making the assumption (yes, it is a big assumption) that he is right, now I wonder which one he was talking about... I'm lousy at detective work

It's pretty obvious that the poster quoted above was referring to F@H. Whether accurate or not, we can't really tell for sure.

Btw, here is another "Anonymous Coward" who claims the opposite:

"A similar approach, Rosetta@Home, led by the most respected protein folding group in the world, The David Baker Lab, uses an algorithm with is much more likely to yield useable results. Last year they used distributed computing to vastly outperform every other protein folding lab in the world."

In terms of scientific contribution, neither of these projects is really worth a great deal. They're both quite speculative, with long-range benefits, if any.

Still, if you have to give one project the benefit of the doubt, it's Folding@Home. Rosetta may "work" better than other ab initio methods, but that isn't saying a whole lot. Ab initio protein structure prediction is just one step above useless, scientifically. A nice game, good for PR and flashy pictures, but the results are usually dismal.

That said, you can still find value in the methods, but only if they give us some scientific insight as to the process of protein folding. The thing is, Rosetta doesn't do that. It can't. It's a heuristic algorithm, and the best you can possibly ever do with it, is say that you're confident to some probability that a particular prediction may be "correct" (in case you're wondering, this probability is usually quite low).

In contrast, Folding@Home is based on a physical model of the world. It may not work very well at structure "prediction," but it's a much better algorithm for other, more realistically-attainable problems (like small-molecule binding). It's also based on actual physical models, and so there is a chance that it can contribute to our understanding of the basic physics of protein folding.

In short: if you truly want scientific relevance, devote your time to other projects. If you insist on helping this field, however, use Folding@home.

(incidentally: I am a computational biologist who has a PhD in the field.)

So there you have another opinion from some anonymous self-proclaimed expert (could be some kid playing with the computer in his mother's basement, as people develop some strange "loyalty" to one project or the other; or even a project person in disguise doing "damage control" against the 1st post you quoted).

Obviously, if R@H manages to do 1.5A RMSD blind predictions for several CASP7 targets this year, then this is hardly "one step above useless", as this guy claims.

On the other hand, if F@H can develop a better method for small-molecule binding than current "docking" (AutoDOCK, LigandFit, THINK etc), it would also be a great achievement.

And then you have the comments from the GlaxoSmithKline guy:

writing that after two years in computational chemistry for what is now GlaxoSmithKline, "I became strongly convinced that computers do not find cures for diseases - or even give you much understanding of illnesses. Molecular modeling is so far from being able to model in vivo molecules that it's practically worthless. ... [W]e already know that trials at this stage are poorly correlated with actual drug usefulness, simulations are just as much a waste of resources as SETI. ... It seems to me that molecular modeling is actually one of those hard 'macho' (but ultimately pointless) projects that gets funding because to criticize it makes you seem anti-drug, anti-therapy and anti-human-progress. (I'm not saying people shouldn't try to model molecules. This is a great blue-sky goal. But people who are trying to find drugs or therapies shouldn't be wasting their time with such techniques.)"

Again, contrary to what the former GSK guy claims, the CureCancer guy from Oxford (doing virtual screening for anti-cancer drugs on grid.org) thought the results were very good by industry strandards.

The screensaver project has produced a large number of 'hits'; molecules predicted to be potential inhibitors of proteins and possible leads for drug discovery. The bottle neck to exploitation of these predictions is the potentially costly synthesis of the compounds and their biological testing.
For two of the series we have passed that stage and the results are very encouraging.
One of the more interesting and challenging targets is the phosphatase.
Much of biology is a balance between phosphatases and kinases. The latter have been important commercial targets for anti-cancer drugs for many years, but the phosphatases have proved more difficult as for one thing the biding site into which the drug must bind is rather ill-defined. The project produced some 128,000 hits for this target. We analysed the results and produced a list of 400 good hits of differing chemical types. These were synthesized and tested. Over 40 [~10%] proved to be genuine inhibitors which is very good by industry standards and what is more they are uncharged molecules which are very different from known inhibitors. We are trying to find pharmaceutical companies to take this further.
For a second series where the target is urokinase plasminogen activator [implicated in prostate cancer] we have again had hits synthesized and tested, this time by collaborators at the Arizona Cancer Center. This too has produced encouraging novel active compounds . These results were presented at the American Association for Cancer Research and a poster giving details is being put on the Oxford Chemistry web site

So, the Oxford guys found value in virtual screening using "docking" software to narrow-down the potential small-molecules (drugs).

Confused? Discouraged? Lots of tough questions, if you genuinely care about the science, rather than credits... Not even to mention things we clearly understand, like excessive redundancy of some projects, wasting so much of volunteer resources for no good reason... DC isn't "free", the difference is an additional eur140/yr per P4 24/7 (in my case).

PS: And to loosely quote some frustrated guy at grid.org's forums, "if all life-science projects work as well as this one [grid.org]", then SETI is probably the most likely one to yield results :-("
Best UFO Resources
Wikipedia R@h
How-To: Join Distributed Computing projects that benefit humanity
ID: 19712 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Send message
Joined: 17 Dec 05
Posts: 555
Credit: 140,800
RAC: 0
Message 19724 - Posted: 3 Jul 2006, 3:37:28 UTC

If that's in reference to F@H, then keep in mind that it's still adding to science (they keep publishing papers) and having more information and insight into the misfolding problems that lead to a few diseases will help with better treatment and eventually a cure.

I know that for an unrelated disease like diabetes, that the ancient folks in Greece had a diagnosis for diabetes (if bees were attracted to a bowl of your urine) and a treatment (eating some type of "earth"). In the early 1900s we got injectable insulin. By the 70's they were making mistakes with ways of increasing the body's production of insulin. By the 90's we had genetically engineered bacteria producing human insulin for us. And while being a juvenile diabetic may suck - at least we have a few treatments for the disease.

With more research into the causes of various misfolding based diseases, we improve the chances of a cure, and give more options for possible treatments for those conditions.


For the one claiming to be from Glaxo Smith Kline - where's the bar for computational techniques of determining the shape of proteins? I believe we got close or matched the experimental accuracy of the current technique in proving a protein's shape. Granted, it may have been a protein with a smaller amino acid chain; but our prediction ability for larger proteins is also improving.

4 years ago, I was told that an RMSD of 4 angstroms was the point where predictions started being useful. Has the point of being useful suddenly dropped to an even smaller RMSD?

ID: 19724 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Tarx

Send message
Joined: 2 Apr 06
Posts: 42
Credit: 103,468
RAC: 0
Message 19732 - Posted: 3 Jul 2006, 15:35:12 UTC

Interesting comments & discussion here!

Regarding that comment from ./ (the one that mentioned "I can also tell you that the project is complete and utter crap, from a scientific perspective."), the F@H project leader actually responded (in the F@H forums) about it:

"There is a wide variation in the success of distributed computing projects (and the claims of what could be done made by those projects). Out of scientific courtesy, it's not appropriate for me to critique other projects in this forum, as such discussions are best left for the literature and other science forums.

One can guess which project he/she's talking about above, but from my point of view, it's bad form to anonymously make such claims. If the person making those claims feels so strongly about it, he/she should make the claims in public so the project he/she is criticizing can rebut and so this isn't used as a general smear over the important science that is being done by distrbuted computing in general -- just because one project may not have been successful (or even well designed from the begining) doesn't mean that d.c. has not been successful!

I can happily talk about what we've been able to do. Our list of peer reviewed papers from FAH http://folding.stanford.edu/papers.html is some sense of the scientific significance. There are major journals there (Science, Nature, etc) and results which would not be possible without distributed computing. We have done what we set out to do (fold small proteins and peptides) and have set new goals, with even greater biomedical impact expected.

Also, FAH distributed computing has lead to major awards from science societies. Most recently, from the Protein Society: http://www.proteinsociety.org/pages/page03a.htm
but there are several others, see http://folding.stanford.edu/awards.html
From the Protein Society's award web page: Quote: The 2006 awardee is Dr. Vijay Pande (Stanford University) for his unique approach to employing advances in algorithms that make optimal use of distributed computing, which places his efforts at the cutting edge of simulations. The results have stimulated a re-examination of the meaning of both ensemble and single-molecule measurements, making Dr. Pande’s efforts pioneering contributions to simulation methodology."

ID: 19732 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Keith Akins

Send message
Joined: 22 Oct 05
Posts: 176
Credit: 71,779
RAC: 0
Message 19734 - Posted: 3 Jul 2006, 16:10:25 UTC

In resopnse to one of my questions on the numb3r crunching forum, David Baker said the following:

"After CASP we will start running both docking and design calculations on rosetta@home, as well as continuiing to improve our structure prediction methods."

That is to say that Phase 2 begins after CASP 7.

This is where Rosetta gets the chance to really prove itself. From what I understand, Rosetta still seems like the "New Kid". As such, Rosetta still hasn't gotten the peer review awards and presskits that F@H has received.

That should soon change as the project leaders take us into phase 2.

In the meantime, I plan on kicking back and pouring up a fine tuscan. Cheers!

No comments Please!
ID: 19734 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Dimitris Hatzopoulos

Send message
Joined: 5 Jan 06
Posts: 336
Credit: 80,939
RAC: 0
Message 19737 - Posted: 3 Jul 2006, 18:22:42 UTC - in response to Message 19724.  

4 years ago, I was told that an RMSD of 4 angstroms was the point where predictions started being useful. Has the point of being useful suddenly dropped to an even smaller RMSD?

I mentioned 1.5A because that was the accuracy for T281 in CASP6 2yr ago, according to R's page accuracy

Also, AFAIK, the resolution of "experimental" (Xray) methods is 1.2-1.3 A
Best UFO Resources
Wikipedia R@h
How-To: Join Distributed Computing projects that benefit humanity
ID: 19737 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Send message
Joined: 25 Oct 05
Posts: 102
Credit: 137,632
RAC: 0
Message 19787 - Posted: 5 Jul 2006, 4:27:19 UTC

It costs on average over $2bn to bring a new molecule based (NME? Forgot what exactly it stands for :() drug to the market. This is well outside the budget of a university and almost all non-profit organizations. What Rosetta@home will do is lower that cost as the structure of more proteins will be known without having the cost of experimental testing for it.

Other than bringing new drugs on the market (if you have a condition, would you really care who sold it to you?) it should also make the development of new drugs cheaper. That means the same R&D budget can get more research done.

After all, I'd rather pay $0.10 more per pill and get new and more effective drugs than safe slightly (or rather let my insurance safe slightly) and get less effective or worse no drugs at all.

It surprises me how many people believe the profit from pharma companies ends up in the pockets of the CEO - those couple millions are pocket change compared to ANY expense such a large company has.

I do get the basics of how Rosetta works, but of course I have no idea on the actual calculations - definatly not my field of expertise :p
However, even if it doesn't let us do anything "new" being able to predict proteins faster and (more importantly) at almost no cost would be a huge advance for science.

A company could set up a server farm matching Rosetta@home's power for some million dollars. Let's go big and assign $100,000,000 - or 100,000 top of the line processors, not counting any discounts for such a large order.
With that kind of "small" expense (given the annual R&D spending of billions of dollars for each of the major companies), a massive amount of research could be done in very little time.

I'm all for helping scientists improve the way they do those calculations, if that means drugs can come to the market faster (no matter who releases them) it means people get treatments sooner -> I can't in any way see how that'd be a bad thing.

Even if the research turned out to be flawed, we all learn from mistakes. The sooner we find them, the better. I definatly don't see an argument for shutting your PC down - even searching for SETI is more productive than doing nothing :p
ID: 19787 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Previous · 1 · 2

Message boards : Cafe Rosetta : "Is Distributed Computing being Distributed Badly?"

©2024 University of Washington