We had discussed the way to make Rosetta@Home have a substantial participant base. Just as a note, my unofficial "poll" shows that we have been gaining about 1,000 systems a day and 5,000 more results in progress ...

A couple days ago, I wrote down the numbers off the status box:

In Progress: 	15,687
Success: 	43,623
Hosts:           5,287

In Progress: 	20,474
Success: 	56,818
Hosts: 	         7,197

So, it is not too shabby ...

---

We had discussed the way to make Rosetta@Home have a substantial participant base. Just as a note, my unofficial "poll" shows that we have been gaining about 1,000 systems a day and 5,000 more results in progress ...

A couple days ago, I wrote down the numbers off the status box:

In Progress: 	15,687
Success: 	43,623
Hosts:           5,287

In Progress: 	20,474
Success: 	56,818
Hosts: 	         7,197

So, it is not too shabby ...

It is amazing.

I was chatting with the boss, DB, about all the folks that have joined and we were both surprised. We had imagined that the bulk of the participants would be students here on campus with the odd member from elsewhere. But, as the more DC-savy among you could have guessed, the VAST majority of the participants are individuals from all over the place, with differing interests and motivations.

It is very interesting to wander through the profiles and read the who, what & where. I spent some time this last week working back to your sites and see what you're doing. I have to say that you guys are crazy! The amount of time invested in your hobbies and the work of others makes me feel lazy as Hell.

Anyway, we've been surprised. The initial response makes us wonder how we can make it possible for us to farm out more of the science rather than just the computation....I don't know if folks would be really interested but it seems to me like a large pool of interest and vigor is waiting to be tapped...

---

Anyway, we've been surprised. The initial response makes us wonder how we can make it possible for us to farm out more of the science rather than just the computation....I don't know if folks would be really interested but it seems to me like a large pool of interest and vigor is waiting to be tapped...

We would love to help out on the science, too! There is a great deal of interest and vigor!

---

Regards,
Bob P.

{background: I was chatting with one of the scientists yesterday about the BOINC community and he (J. Schonburn) said that he said that he was going to be talking with the BOINC guys in Berkeley about the possibilities of making the BOINC project more interactive in some way. The question is, how do you spread out parts of the scientific problem w/o having to train folks in some complicated way. So, folks are thinking along these lines.}

Well, I've no idea of any specific issues out there (I gave up thinking some time ago - and have never been happier), but the general sorts of issues that might come up would be typified by the idea of 'similarity' - how similar are two things. This pops up in biophysics all of the time and the typical computational metrics used are really poor (in my humble opinion - IMHO).

Given two protein structures, a predicted structure and the experimentally observed for example, how similiar are they? This is more challenging than you'ld think. Most folks can look at two structures and readily say that they resemble each other over all, but that the prediction is different in this way or that way. But a computer is really terrible at this sort of thing. This is because you have to tell the computer how to look at the two structures and the computer looks at detailed specifics (distances, geometric orientations, etc.).

Take the siuation in which the predicted structure is exactly the same as the observed structures excepting that it is expanded in volume by 5% (not as 'packed' as well). To the human eye the similarity is obvious. To the computer, the distances are just different, the geometric orientations are just different and the programmer has to tell the computer how to 'weight' the differences so as to provide a sense of simlarity. This isn't easy and so it is hard to program a computer to report something like "Well, the predicted structure is close - but no cigar."

There are many approaches towards giving the computer a sense of 'sort of'. For example fuzzy logic but they are not widely used in biophysics. So, what could BOINC do?

Imagine that your work unit produces three different structures of the same 'energy'. The client could (assuming that you set your preferences) then flash the three structures on your screen and ask you to choose the two that were most 'similar' to the eye. I don't know if this, something like this is going to happen, but I think there are things that just about anyone can do much more easily that a computer and we should use you to do it.

---

Imagine that your work unit produces three different structures of the same 'energy'. The client could (assuming that you set your preferences) then flash the three structures on your screen and ask you to choose the two that were most 'similar' to the eye. I don't know if this, something like this is going to happen, but I think there are things that just about anyone can do much more easily that a computer and we should use you to do it.

This is very interesting stuff! It brings to mind some things NASA was doing with having volunteers count craters on Mars, and also I think looking at very busy sky photos to identify near earth objects or something like that.

I remembered they provided a test photo for one to find things, and then the results would be sent to a volunteer overseer who would decide if you had the eye for that kind of visual pattern identification.

So this sounds like a very intriguing idea, and I kind of like the idea of helping supplement the computer with good old human ability where the computer falls short!

---

Regards,
Bob P.

Imagine that your work unit produces three different structures of the same 'energy'. The client could (assuming that you set your preferences) then flash the three structures on your screen and ask you to choose the two that were most 'similar' to the eye. I don't know if this, something like this is going to happen, but I think there are things that just about anyone can do much more easily that a computer and we should use you to do it.

If something like this does happen, can I suggest two different applications?

The first for those that run single or even a couple of machines that can take the time or have the opportunity to do a visual check. The second would be for those that run the project across medium to large networks that cannot do the interaction part of things.

I'm sure things like this would be thought of but... just in case. :)

---

The Clangers Forum

BOINC Wiki

[quote]Imagine that your work unit produces three different structures of the same 'energy'. The client could (assuming that you set your preferences) then flash the three structures on your screen and ask you to choose the two that were most 'similar' to the eye. I don't know if this, something like this is going to happen, but I think there are things that just about anyone can do much more easily that a computer and we should use you to do it.

It should be pretty simple to let the computer do it by imagerecognition. Boinc lets you use the feature called compound application which means you run different apps after each other. So something like this:

1. send 3 structures to the client
2. the imagerecog. apps picks the best 2
3. the "normal" app runs the dataanalysis on these ones.

- Peter

---

_

It should be pretty simple to let the computer do it by imagerecognition.

Now remember, this is just an example I thought up. I doubt this particular approach would be used in real life.

But, I will say that image recognition likely won't help. For example, I take two structures, invert one and rotate about major axis by 35 degrees - I have the same thing but the image is completely different. It is not that the structures would be exactly the same (if they were then the computational metrics would easily find that), but rather that they are simliar only to a degree.

My simple example envisions a dynamically rotatable image for each structure so that the 3D relationships between objects can be compared in multiple orientations.

---

Now remember, this is just an example I thought up. I doubt this particular approach would be used in real life.

But, I will say that image recognition likely won't help. For example, I take two structures, invert one and rotate about major axis by 35 degrees - I have the same thing but the image is completely different. It is not that the structures would be exactly the same (if they were then the computational metrics would easily find that), but rather that they are simliar only to a degree.

My simple example envisions a dynamically rotatable image for each structure so that the 3D relationships between objects can be compared in multiple orientations.

By saying it should be pretty easy - it's not a 5 minute implementation but the software is "out there". Also for 3d objects it is possible but it is best if it can be exported in some kind of xml-format.

I know it sounds like its a walk in the park - but compared to the amount of manhours required to develop the software should be less than having people looking at the structures for each result.

- Peter

-Peter

---

_

Kel,

I don't think you understand how much I would love to do this type of thing (I'm a 4th year biochemistry geek, so this project and PPAH both make me VERY happy). However, not to rain on the parade or anything, everyone would have to have a copy of CHIME (or an equivalent)... last time I checked, you have to register to get CHIME, and though it's a one time thing, it's another step that would have to be done... unless of course you can get an agreement with MDL to bundle the software.

The other thing is that these visual checks would take time, not necessarily a lot of time, but I just worry that the team would spend countless hours getting this really cool user input comparison program up and running, and then after a week or so no one would use it, or it would become a detraction.

Why? Well, if you have certain WUs that require the user input, Rosetta could get stalled waiting for a user response while the owner is out during the day, or sleeping at night, or any of a million things. This would be a waste of everyone's computing power. Next, when people ARE at the computer, they often have a purpose (at least to a certain extent).

What I propose as a happier medium is to have an option in the preferences that allows users to chose if they are interested in doing this comparison while they are browsing the Rosetta website, sort of like how there are sometimes ads between pages in an online newspaper. Every couple clicks, Rosetta would load up some images and ask you to compare them, with another option to say "not now." This could be seen as an "online" WU, which may or may not get credit... I'd say not, personally.

Anyway, I hope my ramble helps this idea go somewhere, and keep up the great work!!!!

---

Questions? Try the Wiki!
Take a look inside my brain

It is very interesting to wander through the profiles and read the who, what & where. I spent some time this last week working back to your sites and see what you're doing. I have to say that you guys are crazy! The amount of time invested in your hobbies and the work of others makes me feel lazy as Hell.

Before ascribing saintly virtues on us... I am disabled, stay-at-home, with only tentative links to reality at times ... so, when well enough, this is what I do as a "full-time job".

I do come from the systems engineering field with BSCS and MSSE so this type of thing is still right down my "ally". Being relatively well off, I can also afford several computers (currently down to 8) ... As science has been an interest of mine ... well ... confluence of time required, resources available, needs and abilities ...

That explains me ... :)

And today, I spent most of the day reading a book on Hitler and Stalin ...

---

At the end of a wu, you have a three dimensional structure, thus you know the x,y,z coords of each atom. Could you not then choose a reference atom in the structure, I don't know, maybe Carbon atom number 1, and then calculate vectors to all of the other atoms, then use a statistical test to compare the population of vectors from one wu to the population from a second? If the image has been rotated or scaled, there would still be correlations if the stats were done in a thorough manner. Indeed, you could probably correct the scaling and orientation using this technique and then present the two now correctly scaled and oriented proteins to a visual pattern matcher.

Many approaches spring to mind here. A rigid definition of a real problem would help. We may not be biologists, but sometimes it takes someone looking at a problem from a totally different approach to make the connection.

Through 20+ years of software development, it has always been a problem getting the users to say what their problem is. They tend to surround their problem with layers of what they already know, and thus put out a definition which already has built in bias and is a step or more removed from their actual problem.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

I could not add a codicil last night to my post ... actually, it is EVERYONE's Wiki... I just write the most as idle hands are someone's workshop ...

As always, lots of room for improvement ... so if you see a mistake don't hesitate to e-mail me ... (top of the first page has a mailing link)...

---