I am quite new to Rosetta and currently using the loopmodel application to remodel and refine some loops. Everything seems to be working fine. However, upon evaluation of the predicted structures I noticed the absurdly high total energies calculated for the modeled structures. When recalculated with the score_jd2 app, I get an energy of 793748.772 for the model and -1069.460 for the native structure.
What's going on here? I'm using Rosetta 3.2 and the following options with loopmodel:
-ex1 -ex2 -extrachi_cutoff 0
-loops:neighbor_dist 6.0 -nstruct 100
Any ideas for a beginner?
Thank you in advance!
What do the output structures look like? Can you identify any locations where there's any structural distortion, or massive steric clashes? One possibility is that the loops you're remodeling are rather constrained, and Rosetta has difficulty finding loop conformations that fit. Another possibility is that there's something off with your loop file definition, and you're building loops in wrong places.
Another thing is that 3.2 is rather old, and there's been a bit of work (especially on the KIC protocol) to improve loop modeling. You may want to try using one of the recent weekly releases and see if they give better results - especially if you try "next generation KIC" (https://www.rosettacommons.org/docs/latest/Application-Documentation.htm... and https://www.rosettacommons.org/docs/latest/next-generation-KIC.html ) You also might want to try CCD loop modeling for the remodel stage, as it tends to do a better job at modeling than the (older) KIC version. The big hangup for CCD is getting the fragments, but if you use Robetta (http://www.robetta.org/) it's easy and relatively fast.
the refined loops are in their correct places and the output structures are identical to the input (asides from the refined loop, obviously) regarding their backbone. I can see, though, that sidechains are repacked for the whole protein, although I did not specify that?
I checked the energy terms of score_jd2 and asides from those listed below, the terms were comparable for input / modeled structure.
SCORE: score fa_atr fa_rep fa_sol fa_intra_rep pro_close fa_pair hbond_bb_sc hbond_sc fa_dun
SCORE: 792837.796 -11682.81 753455.635 7039.224 6976.786 509.901 -153.797 -23.849 -23.761 38239.9 model
SCORE: -1069.46 -8798.924 1837.689 4632.962 26.497 344.396 -192.52 -401.081 -212.404 3196.873 input
Somethings seems to be way of with the lennard-jones repulsive term and the side chain rotamers... Any ideas?
The problem occured consistantly for various proteins - asides from that I will certainly try the CCD loop modeling procedure. Plus, you definately have a point with the version!
If the loops are in generally the right location, I'd try adding the flag "-loops:relax fastrelax" to the commandline. This will do a full atom relaxation of your structure after the loop remodeling. This should fix up things like minor clashes, problems with sidechain rotamers, etc.