By looking at my results page, I see tasks with names of the form:
mega_lrx_seq_score12_rlbd_xxxx_

...and I note that those tend to have 500,000 or more models run, and they also tend to have RSMDs over 1.5. In other words, we ran 5 or 10 times more models and still did not come up with an answer as close to the true structure as we would like to be, or as close as we get for other proteins.

I was curious what is behind these work units? What is being studied with the larger sample size?

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

By looking at my results page, I see tasks with names of the form:
mega_lrx_seq_score12_rlbd_xxxx_

...and I note that those tend to have 500,000 or more models run, and they also tend to have RSMDs over 1.5. In other words, we ran 5 or 10 times more models and still did not come up with an answer as close to the true structure as we would like to be, or as close as we get for other proteins.

I was curious what is behind these work units? What is being studied with the larger sample size?

Here we're intentionally sampling structures ranging from 0 to 20 RMS to learn what the energy landscape looks like. In order to get a better resolution we have to run massive calculation sizes.

---

http://beautifulproteins.blogspot.com/
http://www.miketyka.com/

By looking at my results page, I see tasks with names of the form:
mega_lrx_seq_score12_rlbd_xxxx_

...and I note that those tend to have 500,000 or more models run, and they also tend to have RSMDs over 1.5. In other words, we ran 5 or 10 times more models and still did not come up with an answer as close to the true structure as we would like to be, or as close as we get for other proteins.

I was curious what is behind these work units? What is being studied with the larger sample size?

There are a few reasons to run these large tasks. 1. As Mike said, one of the main aim of these studies is not just finding the structure closest to the native one, but also to explore as much conformational space as possible, so that we can learn whether we can discriminate between the true structure and the wrong ones. 2. To get enough sampling so that we can improve energy functions to better reproduce structural features observed in native structures. 3. To get enough "wrong" high energy structures, analyze them, and explore new protocols to increase sampling efficiency.

As for the sampling amount vs. rmsd, I think it should be noted that for crystal structures, the resolution also becomes poorer with the size of the protein. A 1.5Å resolution structure is still extremely useful for many types of studies. Second, the amount of phase space grows exponentially with the degrees of freedom, which is the length of the sequence. So, doing 5-10 times more sampling may not be that powerful if the protein is even just a few more residues bigger.