You are here

Pairwise decomposable terms in the Rosetta energy function

2 posts / 0 new
Last post
Pairwise decomposable terms in the Rosetta energy function

Dear all,
I've been wanting to try out the Rosetta energy function with DEE/A* but I realize that few of the terms are pairwise decomposable.

I was wondering whether there is a parameter or a simple way to make energy terms like hydrogen bonds and electrostatics (fa_pair) context independent (e.g. make hydrogen bonds dependent only on the donor-acceptor residues).

I've seen some work on DEE/A* using the Rosetta energy function (Fromer and Yanover, Proteins 2009) but it was not clear to me how they decompose the energy function.


Post Situation: 
Wed, 2011-08-03 12:26

If you're doing fixed-backbone repacking/design, hydrogen bonds and fa_pair *are* pairwise decomposable, as the number of neighbors doesn't change unless you move the backbone (this is one of the reason the environment in Rosetta is categorized the way it is). If you don't have a more clever way to implement it, at the very least you could possibly incorporate this information into your DEE process by splitting the residue type (e.g by having a LEU with less than sixteen 10Ang neighbors vs. LEU with more than sixteen 10Ang neighbors, etc.).

Aside from that, you can turn off the environmental dependance of hydrogen bonds. There doesn't appear to be a flag for it, but you can either turn it off with the HBondOptions::use_hb_env_dep() method (accessible through the EnergyMethodOptions option, which itself is available from the scorefunction object), or you could add a line containing just "NO_HB_ENV_DEP" to the top of your weights file.

I can't seem to find a way of turning off the environmental dependance of the fa_pair term in stock Rosetta. (Though if you wanted to, you could fiddle with the core::scoring::PairEPotential class.) Another option is to turn off the fa_pair term entirely, and use hack_elec instead. There have been some indications that turning fa_pair off and using hack_elec at a weight of ~0.5 for all residues is as good or better, at least for some classes of design problems. (Although it hasn't been extensively tested, let alone published yet).

Wed, 2011-08-03 15:14