ROSIE docking2 and symmetric_docking runs both give plots
and spreadsheets containing many columns of data.
Which spreadsheet columns best reflect the quality of a run?
Should the best runs have each plot with a narrow funnel on the left
and a rather wide flat plateau to the right?
Should there be very few points scattered above or below the plateau?
If the plots don't look ideal, how can one improve them?
For docking2 runs (I have been using the local_docking protocol),
I have been trying to improve the plots by taking the best pdb file
from a run, inserting a TER line between the 2 parts I want to dock,
and using the resulting pdb file as an input for a new docking2
(local_docking) run. Is this the best way to proceed? What would
you do instead? Is it better to use the pdb file with the lowest score
or the lowest I_sc value as the initial condition for the next run?
If there is no clear funnel, which pdb file would you pick as the
initial condition for the next run? How will I know when no more
improvement will occur? How will I know when the funnel is as
deep as it can go?
What would you do to improve the results of a symmetric_docking run?
For example, will it let you input an entire symmetric multi-subunit complex
as its initial condition?
What you're looking for in a docking output is to have a "funnel like" score-vs-rmsd plot. This means that both the score and the rmsd for the most "native like" structure is low, and as you get higher in rmsd you get higher in energy. The higher you get, the "more stable" Rosetta thinks the native conformation is, but even with benchmark cases this isn't necessarily a really big difference. Also, the place where you start to plateau doesn't matter all that much. The main thing you're looking for is to avoid dips in the score-vs-rmsd plot at higher rmsd, which indicate there might be an alternative conformation which also docks well. (But keep in mind which structure is the reference for the rmsd plot - you may need to redock/recalculate the rmsd using a different structure as your "native" reference structure.) Additionally, it's really the bottom edge of the funnel plot that you're looking at - it's expected that there will be higher energy structures with low rmsd. The Rosetta energy function is rugged, and it doesn't take much structural change to get a large change in energy.
The "knobs" you have available to you on ROSIE are rather limited. If you were using the full version of Rosetta, I would suggest supplementing the docking runs with constraints derived from experimental data, information from homologs, and your own biological intuition. However, ROSIE docking isn't set up for that sort of runs. (As there are too many variables to give a good generalized protocol.)
I'm not entirely sure on what you're hoping to achieve with the output structure, so it's hard to give a complete analysis of what sort of things you might want to try. (e.g. things you might want to try could vary based on if you are looking to design an interaction
One possibility is to start with alternate starting structures. Slightly different backbone conformations can sometimes result in better docking output. (For example, if a loop conformation is important to the docking interaction, using a different conformation of that loop could give better docking results.) If you had the regular Rosetta release, I might recommend running a relax on the two monomers to generate slightly different conformations of the structures. If you have other version of the starting structures (e.g. other crystallized forms, different chains in the same crystal, etc.) you may want to try docking with them instead.
Another possibility is that the interface you're looking at isn't really the true docking interface. Using the local_docking protocol, you may want to start the two partners with a different interface which might be a better interaction site.
I don't think repeatedly cycling docking results will get you much. The one place where it might is if the docking runs give you alternative minima. Redocking with the alternative minima might find better results in that area. But the sampling should be efficient enough that close output structures (e.g. within 1-2 Ang rmsd to the input structure) won't gain you much extra by resampling them.
For symmetric docking, you're basically limited to changing the input monomer backbone structure. All other parameters are set by the type of symmetry you have. It will only take in a monomer - ROSIE isn't set up to refine a symmetric multimer, although that is something that's possible with regular Rosetta.
Thanks for your detailed response, rmoretti. It helps a lot.