It is always good to study further things that have challenged you in the past. At least that was the reason after CASP8. They were seeing if new tweaks would improve the prediction and how much the prediction would have improved with existing methods if there were more crunching time before submitting their prediction.

---

Rosetta Moderator: Mod.Sense

I'm curious as well.

---

The CASP9 meeting was very interesting.

The general consensus of the assessors this year was that there were a number of groups "at the top", but there were no outright winners-there is a bit of bunching up and many groups did very well by successfully picking from among the models submitted by automated servers (you can imagine this led to a lot of discussion at the meeting). While no group clearly outperformed other groups by a significant margin, the best ab initio structure prediction at the meeting stood out quite clearly. This was a Rosetta@home product entirely-the rosettaServer this year sent jobs out to rosetta@home and then submitted models from the lowest energy clusters. this was without a doubt the best automated ab initio structure prediction in casp.

In contrast to our approach with rosetta@home which carries out a large scale search for the lowest energy structure, other groups have developed methods that cleverly exploit information present in the protein structure database, and with the increase in the size of this database, these methods keep working better. I am now very interested in using such information to help guide the search for the lowest energy structure (which is very hard for large proteins), in the same way we have been using experimental data to guide search. Some of the jobs going out in the next months on rosetta@home will be testing approaches for taking advantage of this information.

I apologize for the recent system crash. we are beefing up our servers considerably so hopefully this won't happen again for a long time to come.

---

Thanks Rochester for a very interesting article on what the University of Michigan is accomplishing in protein folding!

---

I downoad today some _casp9_ wu. Is not finished casp9 competition??

---

I downoad today some _casp9_ wu. Is not finished casp9 competition??

Yes, CASP 9 is finished, but they often repeat old CASP tasks to see if their new ideas would have produced better results.

Yes, CASP 9 is finished, but they often repeat old CASP tasks to see if their new ideas would have produced better results.

Still CASP9 wus......after 3 yesars!!

---

Yes, CASP 9 is finished, but they often repeat old CASP tasks to see if their new ideas would have produced better results.

Still CASP9 wus......after 3 yesars!!

You can never get enough benchmark sets, its surprising how different protein target are from one CASP to another. I personally pull targets from CASP9/CASP10 and CAMEO (another competition we take part in), to get enough variation.

On a general note, we can get the computer to say almost anything we want it to say, we can adjust our parameters endlessly until we get exactly what we want... but this is called "over-fitting." And parameters calculated for one set might not work for another or a completely new target. The purpose of CASP was to provide a blind test to test our protocols.

But during development, as we wait for the next CASP, we once again risk over-fitting, especially if we use only a single dataset. We thus put aside certain targets (CASP9 etc.) adjust our parameters to work well on one set and test it independently on another set.

So every time you see something running with "CASP9" in the title that means we are testing something new!