I have a question regarding constraints and satisfied constraints scoring.
I'm using constraint files for ab initio modeling but I can't see any attributed constraint's score in the score.sc file. Is there a special command that I need to add to obtain constrain scores for calculated structures? Which one and where should it be added (the website recommendation is not clear to me)?
The question arise because in my log output file I see the program reading the constraints but the number of violation is always 0 despite the constraints being unsatisfied! I’m doubting if the constraints are really being applied during the modeling or not.
Thanks for your attention,
What does your commandline and options file look like? (What program and flags are you using?)
I have noticed that when I add weight to constraints of full atom type (-constraints:cst_fa_weight 10) scores related to the constraints appear in the score file. Otherwise, when there is no wheigth or the weight is applied during the centroid modeling (-constraints:cst_weight 10) no score related to the constraints appears in the score file. Is this normal? If the major search for the structure is done during the first part of the modeling (while considering side chains as centroid) why there isn't a contribution of the constraints to the score?
How to estimate the appropriat weigth for constraints?
I'm using Rosetta 3.5 .
my commandline :
$rosetta/main/source/bin/minirosetta.linuxgccrelease @./broker.args -in:file:fasta myfile.fasta -in:file:psipred_ss2 t000_.psipred_ss2 -broker:setup full_lenght_rig_22_165_tpb -database /path to my /database -frag3 aat000_03_05.200_v1_3 -frag9 aat000_09_05.200_v1_3 -nstruct 20 -constraints:cst_fa_file constraint.cst
My flags :(broker.arg)
## Loop building options
example of constraint file:
AtomPair CB 51 CB 221 FLAT_HARMONIC 15 15 1.0
AtomPair CB 55 CB 220 BOUNDED 10 20 0.5 Experimental_constraint
That does sound reasonable. There's two things that are needed when using constraints with Rosetta. The first is that you need to add the constraints to the structure, and the second is that you need to enable the constraints with a certain weight in your scorefunction. The other thing to know is that there's two type of constraints in Rosetta fullatom constraints and centroid constraints. They're very similar, but they're applied separately. Centroid constraints are only active during the centroid low resolution stage, and full atom constraints are only active during the full atom refinement stages. Turning on centroid constraint weights (which is what -constraints:cst_weight 10 does) won't be applied during the final score of a full atom model, as it isn't used for full atom scoring. It will affect what model you get during the end, though, as it will change the sampling during the low resolution stage. (Assuming you told Rosetta to also add constraints to the centroid structure.)
The main reason for the split is that centroid mode doesn't have all the sidechain atoms (or apolar hydrogens). So you have to approximate sidechain constraints with backbone constraints for centroid mode. If you only have backbone constraints (or CB, which is also present in centroid mode) you can use the same constraint file for both. So use both -constraints:cst_fa_file and -constraints:cst_file and both -constraints:cst_fa_weight and -constraints:cst_weight to enable constraints throughout the whole of the protocol.
As far as what constraint weight to use, it depends highly on how confident you are in your constraints. The more highly you weight the constraints, the more Rosetta will rely on the constraints when making the structure. The lower the weight, the more it relies on the standard scorefunction. You're looking for a balance, where you put on the constraints highly enough to bias the sampling, but not so high you get artifacts from potentially inaccurate constraints. I might recommend trying a range of constraints, and seeing how the structures look. If the resulting structures aren't satisfying the constraints enough, increase the constraint weights. If the resulting structures look unnatural or contorted, decrease the constraint weights. You also might want to try plotting the score of just the constraints (adjusted to weight=1.0) against the score without constraints. You're looking for a constraint weight which simultaneously gives good constraint scores and good without-constraint scores. It'll probably be somewhere in the 0.5-10 range, but it depends somewhat on your system. (I'm guessing it'll also be broad, so you don't necessarily need to do reoptimization for everything.)