When submitting a sequence to the robetta Structure Prediction server I get several structures back, I would like to know these structures are the results of how many decoys? 25,000? 1,000,000? is the final result clustered? does robetta run ab initio the way we do on an HPC?
any details would help.
It's kinda hard to say how many decoys are actually produced, as from what I understand Robetta often farms out much of the abinito runs to Rosetta@home, so the actual number of structures returned is highly dependent on running conditions of the Rosetta@home users. But from what I can tell from the settings I see, the target number of decoys is in the 100,000 range.
The post processing is along the lines of what's recommended in the Baker lab CASP papers, so you will have sorting by score and clustering to remove decoys which are similar (so you have a range of structures in the top 5 models).
The Robetta ab inito runs should be more-or-less what you would do on an HPC, although many of the preparation/post-processing steps are automated for Robetta, whereas you would need to put them together manually for a run on your own cluster.