Gamers Beat Algorithms At Finding Protein Structures 80
jamie writes "Researchers have turned the biochemical challenge of figuring out protein folding structures into a computer game. The best players can beat a computerized algorithm by rapidly recognizing problems that the computer can't fix. From the article: 'By tracing the actions of the best players, the authors were able to figure out how the humans' excellent pattern recognition abilities gave them an edge over the computer. For example, people were very good about detecting a hydrophobic amino acid when it stuck out from the protein's surface, instead of being buried internally, and they were willing to rearrange the structure's internals in order to tuck the offending amino acid back inside. Those sorts of extensive rearrangements were beyond Rosetta's abilities, since the energy changes involved in the transitions are so large.'"
Great idea but seems tough to gamify problems (Score:5, Interesting)
I thought Foldit was actually a pretty fun game and a great idea when it came out, and now that I'm reminded of it I'll have to go back and play some more. It's fantastic to have validation that humans are still excellent pattern recognition engines compared even to very modern algorithms and powerful computers.
But to extend the idea more generally, seems rather hard. Foldit had the great insight to take you to an algorithmically close starting place and let you complete the final adjustments - in that way the algorithm itself is as much a part of the team as the detail or adjustment members they were talking about.
I wonder how many other ideas can be so easily brought to a place close enough that a human can recognize patterns enough to be of use in a final solution. I look forward to seeing what astronomers come up with...
Re: (Score:2, Interesting)
While not quite a game, astronomers already take advantage of semiautomated human pattern recognition: http://www.galaxyzoo.org/ [galaxyzoo.org]
Very cool but needs better clients (Score:4, Interesting)
That was exactly the kind of thing I was looking for to see how the ideas would translate to astronomy...
However the clients (or at least the client I tried) are not great. One of the nice things about Foldit was the UI for manipulations was really well thought out and made it easy to manipulate a pretty complex 3D object, also easy to undo flawed changes. In the galaxy matching game at the link you, had, I got one galaxy pair close to a match but one of the galaxy had spiral arms reversed from the real image, that I could not figure out how to correct for - and then after I clicked on "mass" the whole thing became an oval instead of a spiral, and would not revert no matter how I adjusted things.
I hope they are seeking some funding to expand work on clients for that because they could get some useful analysis from that I think.
Re:Great idea but seems tough to gamify problems (Score:5, Interesting)
"e. It's fantastic to have validation that humans are still excellent pattern recognition engines compared even to very modern algorithms and powerful computers."
Computers primary advantage is speed, for all our pattern recognition capability are mathematical capability is pretty limited besides modern computers. I think it's our ability to stitch or see things as wholes instead of millions of unconnected parts that gives us an advantage - we can recognize things like context that speed up the process significantly whether we are consciously aware or unconsciously recognizing context.
Re: (Score:2)
Imprecision is our main advantage. If you want a computer to fold proteins you either need to program every rule and every thing it will encounter into it or give the computer some sort of guideline-laws that let it figure out things based on their objective properties (probably much more difficult to code and expensive to run).
With a human you just create a UI that doesn't let them do anything impossible and lets them physically manipulate the protein as though it were a physical object with transformers s
Re: (Score:2)
"Imprecision is our main advantage"
I don't think you mean to say this, I think you mean making arbitrary models or metaphors and remapping them onto new scenarios is what we do best. Most of what we do is taking things we've already learned and mixing them and matching them in new ways, I doubt imprecision is our advantage since our ability to sift what details matter from what details don't given our limited computational ability is the whole point if context. i.e. you don't have to go through every litt
Re: (Score:1)
Those sorts of extensive rearrangements were beyond Rosetta's abilities, since the energy changes involved in the transitions are so large.
A human, on the other hand, can just say, "To hell with the energy, I'll just do it anyway."
Re: (Score:2)
Re: (Score:2)
Or perhaps what he really meant was "imprecision is the result of our main advantage," our main advantage being the ability to very quickly get the gist of what is going on and not get bogged down in minor details that might not matter. This would have been important for our long dead ancestors when they ran into animals higher on the food chain. While a general assessment would be appropriate (is it a tiger cub or a full grown male), the sex, exact weight, length, and striping of the tiger would typicall
Re: (Score:2)
Computer's primary advantage is that it was custom-built for the task, rather than evolved into one task and adapted to another. A human brain has at least a hundredfold advantage in raw processing power, however abstract thought - including mathematics - runs over layers and layers of interpreted virtualization (think Python running a JVM running an x86 em
Re:Great idea but seems tough to gamify problems (Score:4, Informative)
Foldit had the great insight to take you to an algorithmically close starting place and let you complete the final adjustments - in that way the algorithm itself is as much a part of the team as the detail or adjustment members they were talking about.
This kind of problem solving was suggested already by Stanislaw Lem in his book Summa Technologiae [wikipedia.org] as a kind of "augmented intelligence" as opposed to purely human or purely artifical intelligence.
Re: (Score:1, Redundant)
http://www.galaxyzoo.org/ [galaxyzoo.org]
As an added bonus, you get to look at some neat deep space photography.
Re: (Score:1)
it is tough but done right it can yield great results. i think this simple crowd sourcing of science wild be one fo the best moves for humanity. you may not have a background in a particular field but still see something that the 'experts' missed.
games are a great way to do it too. if its fun people will put tremendous amounts of time an effort into it. imagine if WoW had a deeper scientific undertone or other learning aspect.
personally i had an idea for a programming language RPG. i realized i could memori
Re: (Score:2)
I am intrigued by your idea and would like to subscr...
No seriously, that actually sounds like a really interesting idea. Perhaps someday you can have time to more fully develop that!
Re: (Score:1)
Yeah, great idea. (Score:1, Insightful)
Teach AI to copy human's best behaviour and create your own replicants.
Great idea, indeed.
Re: (Score:2)
Would that be paranoia or dreaming?
Re: (Score:2)
I've played a bit (Score:1)
Re:I've played a bit (Score:4, Insightful)
While technically true, pattern recognition is the core of our intelligence.
More accurate to say that your intelligence lays in other areas than spacial recognition.
Re: (Score:1)
While technically true, pattern recognition is the core of our intelligence.
More accurate to say that your intelligence lays in other areas than spacial recognition.
thank you.
Re: (Score:1)
that's okay! :)
forget about Foldit! Just download Folding@Home and let your CPU/GPU do it for you!
http://folding.stanford.edu/ [stanford.edu]
Re:I've played a bit (Score:5, Informative)
forget about Foldit! Just download Folding@Home and let your CPU/GPU do it for you!
FoldIt and Folding@Home are doing completely different things. FoldIt (or more specifically, the Rosetta software underneath it) is attempting to guess the final structure of novel protein sequences, using a variety of clever tricks such as mining the database of known structures for peptide motifs. It contains energy functions to evaluate candidate structures, but it is not simulating physical processes, and it tells you nothing about how the linear chain of amino acids forms the 3D structure. Folding@Home is used to study the process of protein folding, where the end result is already known; it isn't useful as a structure prediction tool. Both programs require a massive amount of computing power, but for very different reasons. Both are very useful, but there is almost no overlap in their practical applications. (And it should go without saying that while they can both be an excellent complement to experimental studies, neither can replace them.)
Re: (Score:1)
In 10 years... (Score:2, Informative)
Re:In 10 years... (Score:5, Insightful)
Re:In 10 years... (Score:5, Funny)
I am pretty sure I already do this.
On the plus side, if I can just plug in, it could reduce the eye strain from staring at a monitor all day.
Re: (Score:2)
Re: (Score:2)
Isn't eye strain caused by muscle fatigue? Also, I didn't realize you could have strained nerves.
Re:In 10 years... (Score:4, Funny)
Re: (Score:2)
you're skipping many abstraction layers:
- video RGB data
- screen pixel data / electric levels
- liquid crystal polarization/
- light emitted from screen, filtered through liquid crystal
- light transformed through eye optics
- chemical levels in eye sensors
- neural impulses.
A lot of lossy transformations between different formats. We lose color on LCD. We lose field of vision on screen size. We lose sharpness on eye optics. We lose resolution and framerate on eye cones. The neural display has a potential to be
Prometheus (Score:2, Funny)
Hopefully someone will crack that Ancient code any time now, and we can finally find Destiny!
Makes you wonder... Did the foldit guys borrow the idea off TV, or did the TV writers borrow it off them?
Re: (Score:2)
Or did they just read Ender's game or...
Not exactly a new concept.
Re: (Score:2)
Hopefully someone will crack that Ancient code any time now, and we can finally find Destiny!
Makes you wonder... Did the foldit guys borrow the idea off TV, or did the TV writers borrow it off them?
The Last Starfighter [wikipedia.org], and get off my lawn.
Gamers? no Nerds? yes (Score:2)
Do they really think that gamers are playing FoldIt?
Re:Gamers? no Nerds? yes (Score:5, Interesting)
Re: (Score:2)
Probably because the researcher who wrote the algorithm was bad at representing the problem abstractly, which in turn was because they didn't recognize the connection between that subdomain (protein folding research) and the rest of scientific knowledge.
Re: (Score:2)
Rubbish, if you critsize the original algorithim for being an incomplete solution then you must also critisize the bot for being an incomplete solution. The bot programmer is using the original algorithim's output as his input. The two algorithims are solving different sub-problems and only when you run them in series do you get a better answer.
Re: (Score:2)
Re: (Score:3, Insightful)
Re: (Score:2)
That sounds like a result in itself. By packaging the problem as a game, he's managed to enlist free help from random people on the internet - crowd-sourcing, if I can use that awful buzz-phrase. Perhaps its a victory for scientific marketing more than science, but a better algorithm is a real result that speaks for itself.
We also shouldn't be surprised that an extra few thousand people looking at the problem can come up with a better algorithm than "the professionals". Maybe the researchers were too close
Re: (Score:1)
Obviously. Gamers are used to 'protein structures' (Score:3, Funny)
Re: (Score:3, Funny)
Score: -1 Eeewwww
Curious (Score:1)
You'd think gamers would rather blow things apart than put them together.
Re: (Score:2)
some of us love 3d space puzzles. even something as simple as "fold paper like so, punch holes, now pick the correct unfolded paper" will entertain me for a while. It's better if the puzzles are harder, like that MC escher game on the ps3. http://en.wikipedia.org/wiki/Echochrome [wikipedia.org] I have a feeling that once fold.it comes back online after the slashdotting I'll be able to waste hours folding things.
Congrats, you might already be a Nature co-author! (Score:4, Informative)
Seth Cooper, Firas Khatib, Adrien Treuille, Janos Barbero, Jeehyung Lee, Michael Beenen, Andrew Leaver-Fay, David Baker, Zoran Popovi & Foldit players
So if you've played Foldit you have helped with the authorship of this paper. Not only that, but since it is a biological paper, you are a corresponding author (by virtue of being the last name on the list).
I would highly recommend listing that on your CV, or at least in your application to the Nobel Committee.
Re:Congrats, you might already be a Nature co-auth (Score:4, Funny)
Re:Congrats, you might already be a Nature co-auth (Score:4, Funny)
Re:Congrats, you might already be a Nature co-auth (Score:5, Informative)
If the Andrew Leaver-Fay in this paper is the same one who cowrote "Faster placement of hydrogens in protein structures by dynamic programming" with Jack Scott Snoeyink, then yes. It gives a 4.
Best shot since I missed out on getting a 2 on Ebay.
Re: (Score:2)
That's odd (Score:2)
I figured gamers were better at finding carbohydrates.
Cheetos.
foldit can't be just a game (Score:1)
This causes many people to give up on foldit after a short while, because it takes time to learn what gains points. What is cool about the game is that many of the best players kno
Bio chemistry question (Score:2)
Can somebody knowledgeable about this explain quickly: what makes the real proteins not get "stuck" in the local energy minimums that the program keeps getting stuck in?
Re: (Score:3, Informative)
Re: (Score:3, Informative)
Re: (Score:2)
Duh. There is no need to post stuff that we all know already. ;)
Re: (Score:2)
You have asked the million dollar question. I will give you the pithy answer first; the computer models are inadequate. Real proteins are basically in constant motion; "breathing" and sampling alternative conformations quite rapidly. This is important, first off our cells recycle proteins as a form of regulation; it would be to our disadvantage if all proteins were as rock stable as say, collagen. Foldit also doesn't fully model this bewildering kinetic sampling nor the effects of salt, solvation, etc.
As fo
Re: (Score:2)
As a corollary to the parent, many proteins require chaperones to help fold correctly. The chaperones interact with the protein as it is being translated from the mRNA and help guide it to a folded state. Of course, we don't fully understand any of it.
Re: (Score:2)
And if you're still listening, folding even starts on the ribosome before the protein is fully translated. The "relevance" of folding experiments performed in isolation are always of questionable reality.
Re: (Score:2)
what makes the real proteins not get "stuck" in the local energy minimums that the program keeps getting stuck in?
Several good answers below, but there isn't really a single simple one. This is called Levinthal's paradox [wikipedia.org], and it's one reason why ab initio molecular dynamics simulation turned out to be a terrible way to predict 3D structure. It turns out that Folding@Home (not FoldIt) requires petaflops of computing power because of the same problem: it runs ab initio simulations, and most of them get stuc
beware the music (Score:2)
Re: (Score:2)
You might want to try an anger management course, or something.
Wish we had used this in class. (Score:1, Interesting)
Good lord. Why couldn't we have used this in Biochemistry class?? Just last year they had us using this horrid program called 'O' as the official modeling software for one of our sections. There were some specific questions that had to be answered from the 'O' perspective, but everyone was switching back and forth to jMol so that we could actually see something informative..
I know I know, when you were young molecular modeling meant hand-crafted stick figures. I'll get off your lawn now!
The title is WRONG (Score:2)
The article is horribly misleading - it suggests that the FoldIt users came up with better structures completely on their own. What they were actually doing was improving the automatic structure predictions. If you actually read the Nature article, it shows comparisons of the initial model, final model, and experimental structures. The initial model has the overall fold correct, but with some gross errors which the FoldIt players corrected, leading to an even better model. This is still an impressive re
WOW should involve protein sequencing (Score:1)