Any word on CASP 11? Will there be one in 2014 or have we progressed beyond that?

---

http://predictioncenter.org/casp11/index.cgi

That link says May-August 2014.

---

http://predictioncenter.org/casp11/index.cgi

That link says May-August 2014.

Thanks! I was also wondering if Rosetta had any special goals or plans to participate.

---

Hi Ed,

Yes, Rosetta will be participating at CASP11. We've been working on improving the protocol. I'll post more details coming closer to May.

-krypton

Hi Ed,

Yes, Rosetta will be participating at CASP11. We've been working on improving the protocol. I'll post more details coming closer to May.

-krypton

Thanks! For we geeky competitive types, this is like the Superbowl!

---

I'm crunching _casp10_ wus.
I think rosetta team is preparing to race (re-run old wus with new app)!

---

Not sure if you guys seen all the jobs with "krypton" in the name and the word "cst" somewhere in there. These are some test jobs I am running in preparation for CASP11 that starts May 1st!!

The cst = constraints. We are using sequence covariance information to restrict the search space, by adding constraints into rosetta. If you wanna learn more about what this is, check out my intro on this page here:

http://gremlin.bakerlab.org/gremlin_faq.php

Any questions/comments are welcome =]

Hi.

It's the 2 of May now, I haven't sighted any CASP11 tasks yet, unless there called something else?

---

P . P . L . we are running many CASP11 jobs, though they are being submitted through our automated server which assigns a ID based on DATE of submission to it =[ So it maybe difficult to spot any specific CASP target.

But if they start with "rb_" (indicating [r]o[b]etta server) it is most likely a casp target, since May 1st we've only been using Robetta to submit casp jobs!

=D

Hi krypton.

I've been getting a few of these sort of jobs, guess they could be the ones you mean.

rb_05_05_48197_93921_ab_stage0_t000___robetta_IGNORE_THE_REST_06_05_164071_58_0 finished

Starting task rb_05_05_47197_92508_ab_stage0_t000___robetta_IGNORE_THE_REST_03_19_163460_93_1

Thanks.

---

Hi P.P.L,

here is how to parse the "rb" jobs,

rb_05_05_47197

rb = [r]o[b]etta
05_05 = May 5th
47197 = robetta id

Take a look here for the robetta queue:
http://robetta.bakerlab.org/queue.jsp

In this case, this is not a casp11 job, but it's job submitted by a scientist to the robetta server. I was wrong, it looks like we are still processing other jobs, but casp11 jobs have the highest priority!

Hi krypton.

Thanks for the link, very interesting.

---

Just in case anyone is curious, CASP 11 is almost complete!

If you see any jobs with a prefix Tp/Tx/Tc/Ts these are the casp targets.

This year the CASP organizers decided to change things a little.

For each target amino acid sequence, there are 4+ stages, at each stage more and more information is provide. What the organizers are interested in determining is how much and what kind of information is needed before we can predict the structure accurately.

T0 = no information except the sequence is provided
Tp = predicted contacts are provided (typically most are incorrect)
Ts = sparse simulated NMR data is provided
Tx = Cross-link data is provided
Tc = Only the correct contacts from Tp are provided.

I can go into more detail about any one of these, if anyone is curious.

At this stage, for some of the targets, we know how far we are from the true or "native" answer, but we won't know how the other groups did in comparison till a few months from now.

Thanks for an excellent easily understandable description. I am curious about each CASP.

Just in case anyone is curious, CASP 11 is almost complete!

If you see any jobs with a prefix Tp/Tx/Tc/Ts these are the casp targets.

This year the CASP organizers decided to change things a little.

For each target amino acid sequence, there are 4+ stages, at each stage more and more information is provide. What the organizers are interested in determining is how much and what kind of information is needed before we can predict the structure accurately.

T0 = no information except the sequence is provided
Tp = predicted contacts are provided (typically most are incorrect)
Ts = sparse simulated NMR data is provided
Tx = Cross-link data is provided
Tc = Only the correct contacts from Tp are provided.

I can go into more detail about any one of these, if anyone is curious.

At this stage, for some of the targets, we know how far we are from the true or "native" answer, but we won't know how the other groups did in comparison till a few months from now.

---

We got some pretty cool results in the T0/Tp category (using predicted contacts based on co-evolution data). I'll post a detailed analysis sometime mid december (thats when we get the official results from the CASP organizers).

Thanks to all the crunchers!

Because of this we decided to go forward and make predictions for many other unknown proteins with similar kind of data. For these the jobs the structure is unknown, though the models should be immediately useful for other scientists studying these proteins!

The jobs id(s), will look sometime like this:
TTDA_cst__t000___krypton_SAVE_ALL_OUT_03_09
UXAA_C_cst__t000___krypton_SAVE_ALL_OUT_03_09

the first four letters are the gene id.

An update on the results of CASP 11 would be appreciated.

An update on the results of CASP 11 would be appreciated.

Casp11 Abstracts

---

An update on the results of CASP 11 would be appreciated.

Casp11 Abstracts

Are you pleased with the results? any surprises? Is CASP still worth dong?

---

Hi all,

Sorry for the delay. Here is the report as promised:

For CASP we had two groups.
- BAKER-ROSETTASERVER (no human intervention, Robetta server)
- BAKER (human intervention)

For some targets that were "easy" only servers could participate. If the target is "hard," human intervention is allowed.

For each target five models could be submitted. Practically (outside of CASP), when a scientist gets 5 models and they differ, he/she can design experiments to test which is more likely to be correct.

If we evaluate based on model 1 only, Robetta server is ranked #5, if all 5 models are considered we are ranked #1!!!

The CASP organizers were really surprised. The reason for the a jump was that we ranked our models based on how conservative the method used to produce the model. Model one was always the most conservative answer (minimal sampling from starting homolog), while for the other models we did lots of sampling using Rosetta@home abinitio folding. Historically, the latter didn't help that much, but this year we had a secret weapon! =P


We used co-evolution data in combination with Rosetta@home abinitio folding.
See the first question here (for more info on co-evolution).
http://gremlin.bakerlab.org/gremlin_faq.php

For the "Human" Group (BAKER), the results are similar, and we once again were #1 if we look at all five models.

On average we produced the best models, thanks to all the crunchers!

There were two targets for the human group that we did a little TOO good on, to a point that the organizers thought we may have cheated O_O. I was invited to give a talk on these... but unfortunately due to timing I was not able to go. But we made a video instead [some people in our lab have a too much free time on their hands] =P

http://www.youtube.com/watch?v=N7Bg9Z9JA60

I guess now you guys know what I look like. =]

===========================
INTERMISSIONS (more results to follow in a bit)
===========================

As mentioned earlier:

For each target amino acid sequence, there are 4+ stages, at each stage more and more information is provide. What the organizers are interested in determining is how much and what kind of information is needed before we can predict the structure accurately.

T0 = no information except the sequence is provided
Tp = predicted contacts are provided (typically most are incorrect)
Ts = sparse simulated NMR data is provided
Tx = Cross-link data is provided
Tc = Only the correct contacts from Tp are provided.

The results I already shared were from the T0 category. How did we do when more information was provided?

Ts, we were #1!
Tx, we were #1!
Tc, we were #4... (technically #2, since the first 3 "groups" were submission from the same lab but with minor variations in protocol).
Tp, we didn't take part (I worked on the first couple, but thing got a little too busy with Ts, Tx, Tc).

See ranking here:
http://www.predictioncenter.org/casp11/zscores_final_assisted.cgi

For Ts and Tc, There was only one group that could compete with the might of Rosetta@home, this was a group in korea with super computers of their own. On average we did very similar, but there were a few targets we did MUCH better.

One of these targets is Ts777, this is a very large protein. Here our score was 0.7 (out of 1.0), with the next best score of 0.4. So what did we do that was different?

Rosetta is very good with sampling helical proteins (even without any additional information). The problem with helical proteins and Ts data (NMR data) is that there is a lot of ambiguity. The only non-ambiguous data is which part of the protein is helical, but how the helices pack was ambiguous.

So we chopped the protein into two chunks (two domains) and submitted them with ambigious NMR data to Rosetta@Home. Rosetta with your help was able to figure it out! We then docked the proteins and refined them. See image of comparison below:



There were rumors of CASP being cancelled, but with the awesome performance this year, the show will go on! =]

Thanks to all those that donated their computers during CASP.