I know that Rosetta's energies are given in Rosetta Energy Units (REU), which are not real free energy units, but which sometimes correlate linearly with real free energies.
This means that if I have a bunch of dGs given by Rosetta, I compare them and interpret these values as Rosetta's telling me that this process is more favorable than this other process, and so on.
I am wondering about the interpretation of a single, isolated, value of dG given by Rosetta. For example, when Rosetta outputs a negative dG, can I interpret this as Rosetta's telling me that this is a favorable process? Or is a single, isolated, dG value meaningless, which acquires meaning only when it is compared with other dG values given by Rosetta?
dG values in isolation have some meaning. Positive energies mean something is "wrong" with a structure (usually a clash). So, a single dG carries some information, especially if it is positive. A negative dG is harder to interpret; it means Rosetta thinks it's plausible. It's just that it's not a great correlation: Rosetta thinks a thing is good or bad, but Rosetta is not always right.
Relative numbers are MORE reliable, but still not perfect. The model with the most negative energy is the one Rosetta thinks is most physically plausible / the dominant member of the ensemble. Simply that a model has negative energy isn't very useful, because only the most negative is "real" in many senses.
An old rule of thumb is something in the range of -2 REU per residue for a "good structure". This varies with conditions, and the scorefunction has changed since that number was divined (empirically), but it's still useful as a rule of thumb. If your structure is only pulling -0.5 REU/residue, Rosetta thinks it still has problems. If you're at -10 REU/residue, you've overfit somehow.
I really like the saying of "-2 REU per residue for a good structure" as my protein is of around -740 REU with 442 residues. Is there a reference (e.g publication) that can support this? Thank you.
You could argue that Figure 1 of Nivon et al. PLoS One. 2013;8(4):e59004. supports the assertion.
Hi R Moretti,
I see that figure now. Thank you.