Hi all,
When refining atomic models into low resolution EM maps (in 3.5-4.5Å range), I often need to use maps that have been sharpened to different degrees for the purposes of chain tracing and identification of sidechain orientations - sidechains are much more obvious in a somewhat oversharpened map, while connectivity is much clearer in an unsharpened or low pass filtered map.
However, I have noticed that when real-space refining a model with sidechains included into an unsharpened EM map, coot seems to be a little enthusiastic in trying to ensure that every sidechain is located within the strongest density - that is, equal weight seems to be give to fitting the sidechains (for which density is very weak in an unsharpened map) as for the backbone (for which density is very strong). This can result in misfitting of both the backbone and sidechains.
This can be partially ameliorated by playing with refinement weights and adding torsion restraints, but I have yet to find a complete solution - long and/or flexible sidechains such as lysine and arginine are particularly problematic in this regard.
An option to ignore or downweight (perhaps using a model b-factor dependent weighting scheme?) non-backbone/CB atoms for the purposes of fitting might be helpful to fix this. Perhaps an option to restrain the helical axis to coincide with the "tube" of strong density at low res might also be good.
Alternatively (or in addition), an option to "fix" or restrain rotamers would be handy, so that once the correct rotamer has been identified (in a sharpened map), the conformation of the local segment can be refined (for example against a map corresponding to a different conformation of the same protein) without degrading the geometry of the model.
Thoughts?
Cheers,
Oli.
|