Google Supercomputers Tackle Giant Drug-Interaction Data Crunch 50
ananyo writes "By analysing the chemical structure of a drug, researchers can see if it is likely to bind to, or 'dock' with, a biological target such as a protein. Researchers have now unveiled a computational effort that used Google's supercomputers to assesses billions of potential dockings on the basis of drug and protein information held in public databases. The effort will help researchers to find potentially toxic side effects and to predict how and where a compound might work in the body. 'It's the largest computational docking ever done by mankind,' says Timothy Cardozo, a pharmacologist at New York University's Langone Medical Center, who presented the project at the US National Institutes of Health's High Risk–High Reward Symposium in Bethesda, Maryland. The result, a website called Drugable, is still in testing, but it will eventually be available for free, allowing researchers to predict how and where a compound might work in the body, purely on the basis of chemical structure."
Which supercomputer? (Score:3)
Re: (Score:2)
> IIRC Google has more of the latter and fewer of the former.
Yes, you are correct. More details here:
http://research.google.com/university/exacycle_program.html [google.com]
"The best projects will have a very high number of independent work units, a high CPU to I/O ratio, and no inter-process communication (commonly described as Embarrassingly or Pleasantly Parallel). The higher the CPU to I/O rate, the better the match with the system. Programs must be developed in C/C++ and compiled via Native Client. Awardees will
Re: (Score:3)
they could use BOINC too -- its the largest supercomputer on earth :)
Re: (Score:3)
The fastest supercomputer (http://en.wikipedia.org/wiki/Tianhe-2) runs at 33.86 petaflops, while Boinc currently runs at 8.26 petaflops (http://boincstats.com/en/stats/-1/project/detail).
3 years ago, Google's computers were used to solve Rubik's cube:
http://cube20.org/ [cube20.org]
We can suppose that combined Google servers are above 33Pflops.
The Oracle (Score:4, Insightful)
These machines are what IBM calls "cognitive computing", they can find answers in data that we didn't know were there. In Watsons case the data is general knowledge (AKA common sense), in this case the data is biomedical. There are lots of video's on you tube about Watson but most are cheesy IBM "what we can do for you" infomercials, the talks from the actual developers and the Jeopardy stunt are worth watching.
I know I keep banging on about IBM's Watson in my posts, and although I contracted to them in the 90's I'm not a shill, I'm a degree qualified computer scientist with 20+ yrs as a commercial developer. I was born the year after sputnik was launched, the technological and scientific progress in my lifetime is unparalleled in human history. I really believe that the "AI" developments we are seeing with HPC today are a revolution like none before, machines that are "smarter" than their creators. It already to the point that no major (physical) engineering project is conducted without the aid of computer models, in fact it's basically impossible to design a modern cpu with first having a modern cpu.
The ancient myth of the Oracle has come to life as a flat screen monitor, it will change everything in a single generation, hopefully in a good way.
Re: (Score:3)
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Would the quality of the predictions go up if they used a more accurate (and thus computationally complex) model of the forcefield? AFAIK DE Shaw Research [deshawresearch.com] is building their "Anton" line of supercomputers for the simulation of molecular dynamics.
True, but they've almost entirely been using simple forcefields (with some of their own improvements); the supercomputer is so they can run the MD simulations for orders of magnitude longer.
Re: (Score:2)
The advantage of (real) Supercomputers is that they can tackle large scale simulations (instead of large numbers of small scale simulations, which is what "Folding@Home" and co. do). Would the quality of the predictions go up if they used a more accurate (and thus computationally complex) model of the forcefield? It would certainly help when it comes to electrostatics: solving the full nonlinear Poisson Boltzmann equation (as opposed to approximations) for all the atoms in the solvent, protein, and drug molecules is a bitch; doing it several billion times during a simulation is ... bitch x 10^9. The problem right now is that predictions from first principles of much simpler systems still tend to get a lot of things wrong: they're really only beginning to get a handle on how water works at this scale.
Re: (Score:2)
Oops
The advantage of (real) Supercomputers is that they can tackle large scale simulations (instead of large numbers of small scale simulations, which is what "Folding@Home" and co. do). Would the quality of the predictions go up if they used a more accurate (and thus computationally complex) model of the forcefield?
It would certainly help when it comes to electrostatics: solving the full nonlinear Poisson Boltzmann equation (as opposed to approximations) for all the atoms in the solvent, protein, and drug molecules is a bitch; doing it several billion times during a simulation is ... bitch x 10^9. The problem right now is that predictions from first principles of much simpler systems still tend to get a lot of things wrong: they're really only beginning to get a handle on how water works at this scale.
Re: (Score:1)
What's the difference? Most supercomputers are in fact a cluster of more or less normal computer parts.
Re: (Score:2)
I do wish they wouldn't be so damn sloppy with language. Google datacenters host what can be described as a supercluster. Supercomputers have much higher interconnection between nodes and much faster interconnects.
Words! They mean things!
Fucking journalists...
Re: (Score:2)
Words! They mean things!
No, they don't—we now live in a post-semantic era.
I found out the other day that "figuratively" is now an accepted definition for "literally". Post-semantic... how else can one describe a situation where $TERM == NOT $TERM? Furthermore, "figuratively" does not currently mean "literally", so the symmetry of equality is broken. This is just madness.
You maniacs! You blew semantics up! Ah, damn you! God damn you all to hell!
Re: (Score:2)
Heh. Yeah, I found that out a few months ago. I got half the office going on the subject. (Rather sadly. Programmers think they're funny when often they're not.)
Re: (Score:2)
Doesn't matter. This task is ideally suited for parallel processing.
Re: (Score:1)
Because the person running it is a pharmocologist and thus that would be their area of interest?
Re: (Score:2)
That would be useless. Those fights are dominated by an argument between those who use the products for profit on the one hand and True Believers on the other.
Re: Why limit it to drugs? (Score:1)
Pharmacology would include any molecule that interacts with the bodies processes.
Re: (Score:2)
Re: (Score:2)
Why limit the purpose of the site to drugs? Can we stick in molecules found in foods?
Probably because (unless you are willing to accept high-school-chem levels of oversimplification, and probably even if you are) computational chemistry is not exactly a problem that has any difficulty consuming all the computational resources we could possibly throw at it (unless you count problems with scaling, which I suspect that it has in spades, just to make computing with cheap interconnect less practical).
Drugs have the virtue of being (relatively) simple compared to most naturally occurring mater
Computer predictions (Score:1, Funny)
Predicting when proteins will dock with targets is fine, but I'll really be impressed if Google's "supercomputers" can predict when users will dock with healthcare.gov.
Re: (Score:2)
Sorry if that offended anyone, but it was intended as a joke, not a troll. (I never know how my jokes will go over here - some have been well received, some less so, but I've never had one marked as a troll before.)
BTW, this one also isn't a troll, it's just an explanation. :-)
"docking", hee hee (Score:2)
http://www.urbandictionary.com/define.php?term=docking [urbandictionary.com]
Yes, I'm fifteen. Admittedly, a rather-well-sexually-educated fifteen...
Re: (Score:2)
A fifteen year old with a six digit UID?
What happened? Your parent's got bored with you and set you in front of Slashdot as a toddler?
The answer (Score:1)
is 42.
The real important questions (Score:2)
I see nobody asked the real important questions.
Are they also checking if you can get high from a substance?
Is somebody going to leak that list?
Re:The real important questions (Score:4, Informative)
I see nobody asked the real important questions.
Are they also checking if you can get high from a substance?
Is somebody going to leak that list?
No need to do that. The Federal Government has gone to great expense and trouble to compile this exhaustive list [usdoj.gov] of drugs that can get all the blinky lights in your brain going.
Re: (Score:2)
Any wanna take bets that they'll find (Score:2)
Satan and reefer madness compounds in cannabis...
For the Results of an Experimental Approach (Score:2)
Re: (Score:1)