Maybe we are getting ahead of ourselves.

What if we talk about the actual PROCESS first.

What is step one, then step 2, and so on.

At each step what is in, what is out.

So, how do we pick a protein to start. Then, how do we generate results, what are those limits.

Once processed, and returned, how is that turned eventually into a graph as shown. I know there is more to it than this.

---

Um, David? Joe? Bob? any one? ...

---

It's extremely frustrating isn't it Paul. We have brains as well as CPU's, but they are not interested!

I have tried at a couple of the projects to get more actively involved in the science but the same attitude seems pervasive.

I think part of the problem is the way academic research is financed. It relies on grants, and grants are more forthercoming to people who are "stars" in the field in that they produce a lot of papers. If the topics were discussed, or even worse, solved, in the public domain, then nobody gets the credit. The fact that science itself has moved forward is not actually the issue.

---

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

We are interested! Just a little busy. I've been meaning to get to this post.

The current tests are comparative modeling, which uses a protein of known structure (solved experimentally and chosen by psi-blast sequence similarity) as a template. The structural information from the template is used as distance constraints. The current test set was chosen to represent a range of different single domain folds and lengths that are not too long or short. A larger set was tested on our local clusters with limited sampling. A subset of these were chosen for BOINC because the initial results suggested that more sampling may help.

So the steps are the following:

1. The above is how we came up with the current set of proteins.
2. The sequence, constraints, and various database files are downloaded onto your computer, along with rosetta. Also, the experimentally determined structure of the actual protein is downloaded and used only for statistics (not for the calculations) which makes it easier for us to analyze the results. If you are interested in the details, I recommend you go to the baker lab website and look at the publications related to protein folding as there is too much to cover here.
3. As stated in previous posts, 10 low-resolution structures are predicted. The human proteome folding project has a good description of this process. The main difference is that we use constraints (information from the template structure). The top 2 scoring low-res structures are then used for high-resolution structure prediction in which the full atomic representation of the amino acid sidechains are used (we call this full-atom relax). Basically, the protein jiggles around searching the energy landscape in a more confined space using a scoring function that considers the full sidechains.
4. When the 12 predictons (10 low-res and 2 high-res) are finished, they are sent back to our server for further analysis (for making the plots). The full-atom scores from the high-res predicted structures are plotted against how close the structures are to the experimentally determined structure (native structure) based on RMSD. The hope is that the score correlates with RMSD so that the lowest scoring structure will also have the lowest RMSD and thus is closest to the native structure.

This experiment will be published. Kira in our lab is going to submit her paper soon which will describe the experiments run on our clusters and on boinc. We will be sure to let everyone know when it is published.