|Anonymous | Login
|2024-02-24 04:40 CST
|My View | View Issues
|View Issue Details
|[All Projects] Input Handling
|Fixed in Version
|0000090: No mechanism to convert centroid silent files to fullatom (particularly for fixbb, which should do this)
|fixbb is the intended executable to use when a user has a centroid structure and wants to make it fullatom: run fixbb and pack everything to get sidechains.
This works fine for centroid PDB input; the PDB reader reads the backbone atoms and adds garbage sidechains which PackRotamersMover then repacks.
This fails catastrophically for centroid silent file input: the silent IO machinery coughs up a centroid PDB, which then causes a crash when packing commences with the regular scorefunction.
|Brian Kuhlman reported this to me.
-fullatom might fix it, even for centroid silent files (??)
Should this be fixed just for fixbb, or somehow for silent file reading?
|No tags attached.
|fixbb most promeniently
|Command Line Used
|fixbb -in:file:silent my_centroid_silent_file.out
|Fixed in SVN Version
rmoretti (Attentive Developer)
I think if you have centroid input, you should get a centroid pose. I believe certain protocols depend on this for silent files, so making the silent file reader always give back fullatom poses with junk sidechains would not work well. (That said, looking at a flag like -in:file:fullatom and acting based on it might be acceptable.)
I'd say that if the PDB inputter always converts a centroid input into a fullatom pose with junk sidechains, that's the bug. The input machinery should leave things as they are, and perhaps there should be a utility function that protocols can call to force fullatom poses, if they want to automagically mock centroid input as fullatom. (Or have the user specify something like -in:file:fullatom if they want to use centroid silent/PDB files.)
The PDB reader (not PDBJobInputter) can read centroid as fullatom as a consequence of its necessary fuzzy matching of residue types. If it sees a valid name3 and valid backbone atoms, it will ignore the CEN atoms and make a fullatom pose. (it minimally packs them by default so the sidechains aren't total junk). Of course, if centroid options are active (I guess in:file:centroid) it doesn't do that and coughs up a centroid pose instead.
I guess you can argue that it's a bug that the PDB reader can noisily convert centroid to fullatom without a specific flag requesting that...but I don't know how that bug-or-feature could be decoupled from the absolutely-a-feature ability to read in PDBs with sidechains that are incomplete in some fashion.
in:file:fullatom and in:file:centroid are hopelessly vague in that sometimes they mean "I HAVE a centroid/fullatom and am warning you" and sometimes mean "I want centroid/fullatom, ignore the nonbackbone atoms of what I give you".
The least-effect bandaid is to throw a SwitchResidueTypeSetMover into fixbb's stack to force fullatom at all times. The widest fix might be to put switches into most of the input machinery to ensure that the pose they return is the type of pose the user wants (preferably specified by some new nonvague flags); while this is possible I think it's pretty unrealistic. I'll take a stab at option 1.
|Note Added: 0000086
|Note Added: 0000087
|new => assigned
|Copyright © 2000 - 2012 MantisBT Group