I would like to start by saying that I am incredibly impressed with the Rosetta team for disseminating the information they have obtained so quickly.

I have several question about the results reported. First, is the RMSD you report for the bond length or for the bond angle?? Are these numbers (2.17 and 1.06) good enough to be published? If not, what kind of range is needed?

Next, are these the RMSD vs the best crystal (ie published results) or vs the native structure? (And how are these calculated?? Do you run the simulation through the crystal defraction program and compare to the native crystal?)

Finally, as Paul mentioned in one of his posts, the program seems to be relatively inconsistant in finding the native structure. How do you think that we can shift the main blob of users down and to the left?? (Obviously easier said than done)

---

Questions? Try the Wiki!
Take a look inside my brain

The RMSD is between the C-alpha atoms of the predicted and the experimentally determined structure (the structures have already been solved experimentally for these test cases, which allows us to test our methods). Below 2 angstroms is within the resolution range of crystal structures in general so it is very good. The 1pvaA data is going to be published. The manuscript is almost ready for submission.

How do you think that we can shift the main blob of users down and to the left??

That is the question we are trying to answer. With the computing power provided by this project from users like you, we will be able to test various methods to try to answer this question.

I thought the numbers looked very good (I just read a paper with a 2.something), but it was listed as simply resolution, and I couldn't remember if they were the same.

Good luck on the paper, when it gets published please be sure to post it on the news as I will certainly be reading it =-)

---

Questions? Try the Wiki!
Take a look inside my brain

On RMSD for work units with unknown native (true) shape

I understand that if you don't know the native shape of the protein (as with CASP targets) you can't calculate (and show) RMS distance.
But you can still set some arbitrary shape as a reference one (for example fully unfolded shape) and then calculate distance with respect to this shape.
Similar structures should still have similar distances to the reference one.
I guess that this could give you better idea on the density/direction of the sampling then if you use just energy function.
Even better, if you could publish energy-distance plot per protein as the results become available (once a day for example) everyone could see which part of the energy-distance space is his/her work unit exploring.

---

On RMSD for work units with unknown native (true) shape

I understand that if you don't know the native shape of the protein (as with CASP targets) you can't calculate (and show) RMS distance.
But you can still set some arbitrary shape as a reference one (for example fully unfolded shape) and then calculate distance with respect to this shape.
Similar structures should still have similar distances to the reference one.
I guess that this could give you better idea on the density/direction of the sampling then if you use just energy function.
Even better, if you could publish energy-distance plot per protein as the results become available (once a day for example) everyone could see which part of the energy-distance space is his/her work unit exploring.

Yes, you are right, and we do this for looking at the distributions in house. we thought it would be confusing to send out an arbitrary reference structure, but perhpas not (?).

---

I don't think that its necessary to send the figure of the whole reference structure (that could be confusing I agree).
But you could print the rmsd of the currently accepted structure and update the energy-rmsd plot accordingly. At the moment this field is left blank (Accepted RMSD ?).
From the information point of view, Searching and Accepted windows are showing more or less the same structure all the time.
If you eliminate Accepted window you could have more space for the energy-rmsd map (then perhaps you could even add some value markers for the min-max energies /distances to the plot).
With updated energy-rmsd plot, user have the map of the local search area...and if you publish aggregate map from time to time, then he/she can see where exactly is he/she searching (or how far away from the natural minimum he was). Map search analogy is quite a powerful one, and most people would find easy to understand the concept of the local - global map.
Next step would be to allow the user to choose the search area (within the constraints set by project team).
Or maybe allow the user to change the step of the resolution search.

---

I don't think that its necessary to send the figure of the whole reference structure (that could be confusing I agree).
But you could print the rmsd of the currently accepted structure and update the energy-rmsd plot accordingly. At the moment this field is left blank (Accepted RMSD ?).
From the information point of view, Searching and Accepted windows are showing more or less the same structure all the time.
If you eliminate Accepted window you could have more space for the energy-rmsd map (then perhaps you could even add some value markers for the min-max energies /distances to the plot).
With updated energy-rmsd plot, user have the map of the local search area...and if you publish aggregate map from time to time, then he/she can see where exactly is he/she searching (or how far away from the natural minimum he was).

we've been talking about ways to do this. we would like to get back to showing users their results in the context of those of everybody else, but this would require some new infrastructure.

Map search analogy is quite a powerful one, and most people would find easy to understand the concept of the local - global map.
Next step would be to allow the user to choose the search area (within the constraints set by project team).

this would be neat as well. it would test a whole new approach to the problem--having human experts inspect a map of the lansdscape built up at a given point and decide where to best focus searching.
might be a great solution to the problem, but how would you do this in a controlled way?


Or maybe allow the user to change the step of the resolution search.

this would be pretty easy to do as there are command line flags that control the resolution.


but all this is for after casp!! we are scrambling to keep up!

---