Medical Research and Computer Power

[coachHinesfan]

Hello there all, I have a question. Why does medical research take up alot of computing power/processing power. For example the folding@home project by Stanford ? Another example being weather calculations or weather research. What exactly makes them take up so much power ? For example, I saw in a news clip that the army just bought alot of PS3's to simulate a human brain. Do all these 3 examples have different traits or something else ?

[jonsca]

Let me preface this by saying that I am not an expert in any of these fields but I have had some exposure to the underlying theory, so facts are AFAIK. Please pardon any textbookiness.

The least common denominator is they all involve the solution of simultaneous differential equations (depending on which model is used). Such work requires large scale matrix transformations, the size of the matrix being proportional to the degree of the equations. The more you can exploit granularity (that is how small of a piece can you parcel out to each individual processor and still be effective) the faster your calculations get done.

There are other specific factors for each case, for example the protein structure prediction involves a lot of combinatorics calculations to posit the pool of possible structures along the way. Weather predictions over long ranges of time require ever so small timesteps to maintain their accuracy. I haven't read anything about the army's project so I'll keep my assumptions on that to myself, but suffice it to say the more neurons you have in the simulation the greater the likelihood of approximating phenomena at the subcircuit level.

Phew, so I don't even want to go back and reread that. :) Hope this gives you some search terms that you can google.

[cwarn23]

Yes and for a medical scientest to do something like calculate the genitic makup of a couples future child that would require huge amounts of cpu as there are billions of dna protines to calculate. So with everything being theoratically possible but the cache is you allways need relavent cpu. And in most cases you need more cpu than what anybody has access to. So in this case to do something like to monitor all of a humans organtic systems to make sure everything is normal then huge cpu can be required as many calculations need to be performed to calculate what is normal for that person. For example, a persons normal weight is factored by age, height and current weight. But more technical organic components require deeper calculations.

[jwenting]

The main problem is the massive amount of data.
The same thing happens in some areas of physics where the amount of data generated in a single experiment can run into the petabytes.

[cwarn23]

The main problem is the massive amount of data.
The same thing happens in some areas of physics where the amount of data generated in a single experiment can run into the petabytes.

Indeed. I wish they would invent those 256 petabite hardrives sooner with 512TB of ram. But soon I believe there is a hard drive that can hold something like 12TB on a cd size disk. Still in development but in another 2 years will be ready for public use with the new replacement of blueray. Imagine that. A dvd that can hold 12TB of video (not GB but TB). So in the near future there will be no problem with space as the disk space technology is growing faster than cpu technology. Just thought I would share that.

[jwenting]

disk space requirements already outpace CP requirements and have for years now.
Unless you're looking to process those petabytes of data in real time of course, but even CERN isn't that ambitious :)

[coachHinesfan]

Hello there all. Thank you for responses. Jonsca, thank you for your response :) however i did not understand a single sentence other than in the beginning paragraph. So far I understand that data is alot. But i still don't understand how that data is reached upon.

Let me give you an example of my view : I understand that making animated films for example Transformers : Revenge of the Fallen will take massive amounts of data because you have to make the 3D models and texture them and do all sorts of things. So for that i understand why it would take days just to render a realistic frame.

But I don't understand how equations can make up/produce so much data that it takes Terabytes to store the info.

[GrimJack]

Well, if we take folding@home as a starting point then we can first look at what a protein is:

Proteins are an important class of biological macromolecules present in all biological organisms, made up of such elements as carbon, hydrogen, nitrogen, oxygen, and sulphur. All proteins are polymers of amino acids.

Proteins are collections of amino acids and amino acids are collections of certain kinds of molecules. There are 20 types of naturally occurring amino acids; these amino acids bind together in chains to make proteins. A simple protein containing 40 amino acids has the possible primary structure composed of 40^20 (40 to the 20th power) possible combinations (not really true as there are rules about how the different amino acids can combine. What the protein is composed of is not the complete picture - there are rules on how amino acids combine into sub-structures - then there are rules how the proteins fold (its 3 dimensional structure - this is what folding@home is working on). There is a 4th level of complexity but you should get the picture that there is a bazillion possible ways to make up 1 small protein then you have to multiply that by a large (but finite) number of ways for sub-structures to form and multiply that by a not so large number of ways to fold the protein.

take a look

[jonsca]

Hello there all. Thank you for responses. Jonsca, thank you for your response :) however i did not understand a single sentence other than in the beginning paragraph.

I apologize. I wasn't trying to talk over your head I just figured I would give you some specific examples and you could Google them to death. For references about the differential equations aspect see http://en.wikipedia.org/wiki/Numerical_ordinary_differential_equations"]this and this but just imagine that there are tens or hundreds of equations that all depend on each other (akin to a system you might have seen in algebra like x -y = 3, x+y = 5). The matrices become huge very quickly.

But I don't understand how equations can make up/produce so much data that it takes Terabytes to store the info.

Imagine you're those physicists at CERN. You want to record the position and trajectory of millions of particles and on top of that you want to take the data at time intervals of 1 ns. Right there you would have a million billion datapoints per second (my numbers are completely exaggerated since I know next to nothing about what they trace but you get the idea).

@GrimJack - thanks for clarifying some of the points on the protein structure prediction.

[jwenting]

Your numbers aren't much exagerrated, except the nanosecond precision (it's more like microsecond) :)

Yes, we're talking about tracking massive amounts of subatomic particles moving extremely rapidly over a short distance accurately.

[Steven Kenneth]

"Because of the high amount of data that is being processed. Let us take the example of weather forecasting. If you've seen the movie TWISTER which came years back, you'd have an idea of how it's processed. Now, that being a movie, you need to understand that it's not only about temperature or the dew point. Its about recording all the metrics all the time. Emphasis on ""all the time"". For predicting the action of, say a hurricane, you need to study hundreds of them inorder to come to a prediction.

Imagine this scenario. You're in your house and you are going to predict the next person who knocks on your door. The parameters we'll be considering today will be person, time, weather.
Person :
Assuming you have a wife and 2 kids, now your predictions would be : Wife knock, kid 1, kid 2, wife and kid 1 , kid 1 and kid 2, wife and kid 2 or all 3 of them. Apart from this it can also be your paper guy, cable guy, next door neighbour, downstairs neighbour or a random passerby who wants to know the address.
Time :
Assuming it's eve, say 4, then may be it's your kids coming back after school. If it's the morning, then the paper boy. If it's at a random time, them could be the courier person or some other random person.
Weather :
Well, assuming there's a heavy thunder shower, any passerby could ask for refuge in your home.
Day :
You'll need to analyse this atleast for a week and atleast for 3 months to predict the behaviour on monthly basis.

Now, when a simple scenario like ""Who'll knock on our door next"" has so many data to analyse, imagine what weather predictions and the said protein folding would have. And for those, one brain won't be enough to handle all those data and so more brains and hence more processing power.

This is a nice infographic which can give you an idea
https://toptrends.nowandnext.com/wp-content/uploads/2017/02/TechBrainMap-170214.png"