| View Issue Details [ Jump to Notes ] | [ Issue History ] [ Print ] |
| ID | Project | Category | View Status | Date Submitted | Last Update |
| 0000232 | Rosetta | [All Projects] Incorrect Results | public | 2013-03-26 12:18 | 2013-04-28 17:36 |
|
| Reporter | tjacobs2 | |
| Assigned To | amelie | |
| Priority | high | Severity | major | Reproducibility | always |
| Status | assigned | Resolution | open | |
| Platform | All platforms | OS | Any | OS Version | Any |
| Product Version | Trunk | |
| Fixed in Version | | |
|
| Summary | 0000232: Kinematic Closure broken with extended loops (and possibly even without them) |
| Description | Kinematic Closure has a severe bug that causes failure to model loops with previously unclosed loops. |
| Steps To Reproduce | 1)Run loop modeling application on a loop that starts out broken
2)Observe that the loop is still broken after a "successful" KIC closure |
| Additional Information | The bug looks like it is due to the implementation of the idealize_loop_first logic in KinematicMover.cc. The code calls conformation::idealize_position(), which does NOT idealize residues at cutpoints in the FoldTree. This is a major problem because then the bond/angles and lengths given to the BridgeObjects function don't match the pose, and are never applied (only the torsions from the solution are applied, even though the set_pose_after_closure function of the perturber takes the bond angles and lengths, it ignores them). Another potentially major problem (which can happen even when starting with a closed loop) is that the perturb_chain() function of the perturber is allowed to modify the bond angles and lengths prior to calling BridgeObjects, but doesn't actually change the pose (it can't, it's given a const reference). This means that, once again, the bond angles/lengths given to the BridgeObjects function don't match the pose and are never mapped back, which can lead to a broken loop.
It's easiest to reproduce this bug using perturb_kic with the kic_leave_centroid_after_initial_closure flag on (which causes this to run very quickly, and also prevents the monte-carlo wrapper around this call from keeping only solutions that happen to be closed).
Probably the easiest way to fix this is to make sure that *everything* returned by the BridgeObjects function (which, in the case of angles/lengths is identicle to what is passed in, which makes me wonder why even bother returning them...) is actually applied to the pose afterwords. |
| Tags | No tags attached. |
|
| Application(s) Affected | LoopModeling, anything else that uses KIC |
| Command Line Used | -s pdb_with_broken_loop.pdb -loops:remodel perturb_kic -in:file:fullatom -loops:fast -nstruct 1 -overwrite -kic_leave_centroid_after_initial_closure |
| Developer Options | Confirmed As Bug |
| Fixed in SVN Version | |
|
| Attached Files | |
|
Relationships 
Relationships