We are actively considering now several different approaches for building interactivity into rosetta@home which I think will be tremendously exciting. In the simplest scenario, you will be able to propose moves to the computer to try out, for example rotating around a bond you select. In the current version, running on your computers, the computer selects moves at random, and then accepts those moves which reduce the energy. In the interactive version, whenever you have an idea, the computer will try out your move instead of picking randomly. This will let those of you who are interested to become actively involved in the searches, which should be more fun than watching the screensaver, and I'll bet there are some protein folding geniuses out there who will be able to work wonders! We are also thinking of having the 5-10 best models found thus far, and the names of the finders, in a panel on the screen that if you want you can start from and try to improve.

---

Rhiju has posted the top prediction we made for CASP7 target 354 on the top predictions page. This is one of the best ab initio structure predictions we made in the CASP7 experiment. Take a look--this is the level of accuracy we would like to be able to achieve consistently starting from amino acid sequences alone. We have been working hard making improvements to the search strategy over the last two months and the next code update you recieve (in the next few days) will have a number of improvement I am optimistic will take us closer to the goal of consistent high accuracy structure prediction.

Also in this next code release is exciting new methodology created by Phil Bradley for modeling the amyloid fibrils associated with many human diseases. Next week Phil will be sending out work units which generate models for the amyloid fibrils associated with Alzheimer's disease. This is not only an extremely important problem from the human health perspective, but is also an interesting challenge for structure prediction--the length of the protein that forms the fibrils associated with the disease is only 41 amino acids, but instead of folding into a single independent structure on its own, it associates with many other copies of itself to form extended helical fibers. Phil has built into the Rosetta code a general treatment of symmetry which allows him to model the folding and simultaneous association of many copies of the 41 residue protein--the constraint that all copies are identical makes it possible to model these many copies without significantly increasing the size of the search space. You will be able to contribute to and observe the symmetric folding and association of the subunits of the Alzheimer's fibril very soon.

---

We are now preparing for the CASP7 meeting next week. Bin and Rhiju will be giving talks describing the work all of you did on both ab iniitio structure prediction and comparative modeling, and I'm going to talk about high resolution refinement a bit more generally. I will give you a full report on the meeting when we return.

Last week I was at a meeting of the Howard Hughes Medical Institute, which supports much of our research. The Institute has just started a new initiative to provide resources to the scientists they support to develop science outreach and education programs. The director of this program was very enthusiastic about our goals for developing the science education side of rosetta@home, both for the general public and for high school students (perhaps as a short unit in science classes). So I hope you will see considerable evolution of the screensaver and supporting materials in this direction over the next six months.

On the interactivity front, a computer science graduate student here, Adrien, just completed the first proof of concept step and is going to give us a demo tomorrow. He will describe his plans on these boards soon.

---

I'm just back from the CASP7 meeting which was very exciting. I will give a fuller report this weekend after I catch up on sleep and the pile of things which has accumulated, but very briefly it turns out that many of the predictions Rhiju and Bin have posted on the "top predictions" page were the best made for these targets in the whole prediction experiment, and for the experts among you, the T283 ab initio prediction was found by the CASP7 assessor to be accurate enough to unambigously solve (by molecular replacement) the xray crystallography phase problem, an absolutely unprecedented result.

---

I'll post more soon on the CASP7 results--for now you can view all of the numerical evaluation of the predictions at the CASP7 web site. The human expert evaluations will be posted there soon as well.

I just recieved the latest edition of the Howard Hughes Medical Institute bulletin which contains an article about rosetta@home participants. You can download this article, and others describing other interesting work supported by the institute from

http://www.hhmi.org/bulletin/popups/download_pdfs_nov06.html

---

Thanks to feet1st for clarifying my original post! His improved version, with an explanation suggested by SOAN, follows:

The assessor reports are not yet posted. In the mean time, you can get an overview of the predictions you all made for CASP7 by looking at this url

Each team participanting in CASP is allowed to submit up to 5 models. You can see each of our submitted models indicated by black lines. The orange lines are the models submitted by some of the other 252 groups that participated in CASP7. Not all groups submitted predictions for all of the proteins.

The y axis is a measure (RMSD) of how close the model is to the true structure for the fraction of the structure indicated on the x axis (from 0 to 100%). Each point (X,Y) on one of the lines indicates that that the best predicted contiguous segment of X percent of the residues in the model represented by the line was on average Y angstroms from the experimentally determined structure. .

A perfect prediction would be a horizontal line at zero. so we would like there to be at least one black line at zero all the way accross for each target (there are five black lines in each graph, one for each of the five models submitted). The first thing to note is that we are not even close to predicting protein structures perfectly! This is why we are continuing to do methods development work as a major part of Rosetta@Home, and we think the predictions would be better if they were made today than these which we made last summer. Thanks again for contributing to our efforts.

How did Baker lab do compared to the 252 other groups participating in the CASP7 experiment? One way to look at this is to count the number of times one of the Rosetta@Home models was clearly better than the models produced by other groups. You could then browse through each of the graphs, and count the number of cases one of the black lines is clearly lower than all of the orange lines for some fraction of the structure.

My son Benjamin and I just did this quickly. Our list is the following:
targets
283
299
299D1
299D2
300
307
316D3
319
323
327
329
330
330D2
331
347D2
350
354
356
357
360
363
365
368
380

From looking at these plots, it does not seem other groups had as many "breakaway" models. If you have a minute, take a look at these plots and perhaps select a different group to see what their plots look like.

Also note that when our predictions are not clearly better then the others, we're often very close to the best predictions. This is indicative of Rosetta's consistently good predictions.

This "numerical evaluation" is only part of the story, and the measure used in these plots only looks at the position of the backbone Calpha atoms, not at the protein sidechains which, as you know, we spent a lot of time modeling as well. I'll give you a report on the expert evaluation and the results for the whole protein chain including sidechains in a day or two.

---

The CASP7 assessor reports are now posted at http://predictioncenter.org/casp7/meeting/talks.html. Take a look at the plot on page 18 of the report on free modelling--congratulations to all of you!

---

Yesterday and today Bill Schief and I are attending the Gates foundation funded
COLLABORATION FOR AIDS VACCINE DISCOVERY KICK-OFF MEETING
I am meeting many of the long time leaders in the vaccine design area
and we are establishing many new exciting collaborations. Computational
protein design methodology has never before been applied to vaccine design,
and there are huge numbers of things to do. For example, yesterday we talked
to the director of the NIH vaccine research center about designing flu vaccines, and
with other experts on viruses about designing Herpes vaccines.

The presentations at the meeting are primarily from the leaders of each of the new vaccine
design projects which the Gates foundation started funding several months ago,
but also from Gates foundation people describing their perspective, including today
a talk by Bill Gates.

---

Today I met with Nancy Hutchinson and her team of educators who work closely with teachers to design
and implement high school science units on proteins and molecular biology. Nancy's team and I are very excited to work jointly together to integrate rosetta@home into these units, and also to develop a "virtual" version of the course that would provide a comprehensive introduction to the science underlying proteins and rosetta@home which would not require a previous science background. Tim Herman, who produces beautiful models of proteins students can hold and manipulate will also be part of these efforts (see http://www.rpc.msoe.edu/cbm/). Nancy's program for teachers is described at
http://www.fhcrc.org/science/education/educators/sep/.

On the research front, the results from the last month of rosetta@home are really exciting. Chu Wang has made very dramatic progress in predicting the structures of protein-protein complexes: previously we were able to predict structures of complexes of two proteins given the structures of the two proteins
solved independently, but only when these structures do not change significantly when the two proteins interact. Now Chu, using the computer power you are so generously contributing, has been able to generate accurate predictions for complexes where the individual protein structures do change upon binding--he does this by simultaneously optimizing both the relative position of the two proteins and their structures; needless to say this is an extremely computationally demanding problem that would not be even thinkable without rosetta@home. This is an extremely important step forward since Chu's approach should now allow the prediction of protein-protein interactions quite generally, which is a very important problem in molecular biology.

Brian Kidd just showed me very promising results on another important problem: the modelling of protein conformational changes. Many proteins change their structure and function when they bind an activiating small molecule. Brian just started running his calculations on rosetta@home, and already has exciting results showing that, starting with the structure of proteins bind to such activators, if he removes the small molecule, the protein switches back to the state it is known to adopt experimentally in the absence of the small molecule. This is an step toward modeling the motions that underlie the signalling pathways used for communication within and between cells.

Finally, our continued analysis of the CASP7 rosetta@home predictions keeps getting more exciting--it turns out many of these blind predictions have an accuracy that is unprecedented, and that allows the models to be used for applications, such as solving the x ray crystallographic phase problem, that seemed out of reach earlier. The CASP7 assessor Randy Read made the first dramatic discovery with Rhiju and Bin's prediction, and he is continuing to work with us to better define the advances that have been made.

The next month or two we are going to start writing papers describing all of this progress which will highlight the contributions all of you have made, and we will also be continuing to attack new problems in the protein structure prediction and design area which you have made it possible to envision solving.
Also, as I've previously discussed, we will be beginning a concerted effort to bring full interactivity to rosetta@home, so you will be able to be involved in the research even more directly in the future!

---