We are currently testing our new methods for modeling membrane proteins and for modeling drug binding to proteins in a CASP like prediction challenge. The challenge is described at:

http://jcimpt.scripps.edu/GPCRDock/default.aspx


This is an interesting and important challenge because this class of proteins, called G-protien coupled receptors, are the targets of a large fraction of current drugs. This particular case is hard, because we first have to predict the structure of the protein, and then dock the drug on to the model. Needless to say, this is a huge computational challenge and couldn't be done without all of your contributions!

---

We are entering an exciting new phase with rosetta@home!

We now know that our casp8 team which used rosetta@home plus some human intuition did very much better than our automated server robetta on the casp8 targets (the full results from all groups will be presented at the casp meeting in Italy at the beginning of december). our goal between now and the casp meeting is to develop a completely automated protocol that does as well as our casp8 team. towards this end, we need to do extensive benchmarking of the new methods that were developed by students and postocs in our group working on casp8 this summer. these benchmarks must be on non casp8 targets, so that after we have found the best combination of methodology we can see how well it would have worked during casp8 without risking building in the answer by "memorizing" the casp8 structures. we tried many different approaches this summer, and testing all the different methods on a large comprehensive set of protein structures would normally be impossible, but with rosetta@home and all of you we can get rigorous statistics on the different methods and rigorously determine which metho is best.

Please read researcher Mike Tyka's posts in his thread:

https://boinc.bakerlab.org/rosetta/forum_thread.php?id=4388

Mike has set up a fantatsic benchmarking system that will let us determine the best methods to use for protein structure prediction. I'm very excited to start using this myself in the next couple of days!

---

We just got good news from the CASP organizers--our results were among the best at casp8 and we have been asked to describe the methods we used -- including all of your work -- at the CASP8 meeting in December. There are a number of evaluations of the performance of the automatic servers at CASP8 available on the web, and comparison of our models to the servers shows that all of your efforts led on average to quite a bit better models. Of course, we don't yet know other "human" groups did on the CASP8 targets--for this we will have to wait for the meeting!

I'm also excited because I've been working on improvements to the energy function which underlies most rosetta@home calculations, and the first tests of the improvements should run on rosetta@home this next week!

---

We have many exciting calculations just starting to run on rosetta@home! We are testing many new methods (see Chu's post today about zinc binding proteins for one of them) and we are eagerly examining each new work unit coming back from your computers. I am very excited personally as I'm doing the first large scale test of (what I think are) improvements in the energy function underlying the searches that I've been working on the past few months (these are the work units with "split" in them).

---

Please see Sarel's post today on protein-protein interface design about to start up on rosetta@home. Sarel is designing proteins to block the action of pathogens like cholera, and we are excited that with your help we can make inhibitors for many proteins causing disease. We are also starting large scale tests of brand new structure prediction methods, including some that I have been developing myself. This is a very exciting time at rosetta@home!

---

Two exciting things today:

First, Rosetta@home is Wikipedia's article of the day for December 1, 2008! Welcome to new participants who learned about our project today!
(we had an unplanned outage this morning, but you can believe we (this means the behind the scenes heroes Keith and David) worked like crazy to get the project back up as fast as possible!).

Second, the CASP8 results are now publicly available! http://predictioncenter.org/casp8/results.cgi.
Your Rosetta@home results are the ones labeled "dbaker" (this is a complete misnomer as I deserve very little credit for any of this). in any event, now you can compare to POEM and other methods, it takes a while to navigate through all of the information now available. If any of you come up with simpler overall summaries please post for the rest of us!

The CASP8 meeting is Dec 3 - 5 and I'll be posting results from the "human expert" evaluators as they come in.

---

Welcome again to new participants!

I can explain a bit how to look at the results at http://predictioncenter.org/casp8/results.cgi:

In the box near the top that says "all groups" click and choose for example "DBAKER" (your predictions at rosetta@home) or "POEM" or whatever method you are interested in. then inspect the plots that appear on the page. each prediction by each group is represented by a red line, and predictions by the group you selected by a black line. the best prediction is represented by the line that is underneath all the other lines (the height is the distance from the true structure, the x axis, the fraction of the protein considered).
your rosetta@home predictions stand out for a number of targets, particular prediction problems like
T487D4 (target 487 domain 4).

---

The last day of the CASP8 meeting is tomorrow--while I was not able to attend James and rob have reported back that the meeting has been very exciting.

there are now summary statistics on the performance of all groups in CASP8 available at

http://prodata.swmed.edu/CASP8/evaluation_human/DomainsAll.Best.Zscore.html#top

---

Mike Tyka's results from your calculations over the past two weeks on rosetta@home are truly exciting! Many of you have found, starting from an extended chain, structures that are close to the crystal structure but much lower in energy. When Mike inspects your very low energy structures, in many cases it seems possible that they may be better models than the experimentally determined structure. We never expected to be in this situation so soon! Our next step is to see whether these low energy structures may actually be as or more consistent with the experimental data than the deposited crystal structure. If true, this would be a new step forward for computational structural biology!

---