To whomever this may concern,
I am trying to run relax with constraints. The constraints are read in, and I verified by hand that the values within the score output file are accurate. However, the relaxation does not seem to move the atoms to reduce the constraint score. The command I am using is relax.linuxgccrelease @inputs.inp, where inputs.inp contains the following text:
and constraints.cst contains a single constraint: "Dihedral CA 1 C 1 N 2 CA 2 BOUNDED 10.0 15.0 0.3 0.5 TAG". I am not sure why the application is not driving the dihedral angle to be closer to the designated range. I am not expecting the constraints to be perfectly met, but the current dihedral angle is substantially different than the designated range and changes by less than a degree during the relax application. Any help would be greatly appreciated.
The first thing to try with constraints is always to pass a bogus constraint and check that it crashes - just include a garbage line to see if it fails as expected. (You've implied that you know they're being read in, so this is moot).
The next thing to try here is to turn the constraint weight up. Your weight is 1, the multiplicative identity, so it won't affect the (raw constraint) score. Try a score of 10 and see if things change. Then try 100. You can also try increasing the "sd" value (0.3).
Finally, if it is refusing to alter the N-most torsions, but other torisons are constrained as desired, then there may be a bug or un-optimization due to the fact that you are very close to the root of the AtomTree. Changing those torsions causes the whole rest of the protein to swing around in space. If there is some other rigid body present (is this a monomer?), then clashes caused by this movement will prevent the minimization you want. The fix here would be to put the AtomTree root somewhere else. This is not trivial but I think we can do it with RosettaScripts.
Thank you very much. Increasing the constraint weight seems to have worked.