On Wed, Jan 7, 2009 at 1:54 PM, Jacob Keller <[log in to unmask]> wrote:
> I am sure that most here have dealt with the issue, when making superpositions of conformationally-different structures,
> of which regions to align as references and which to call "mobile." Conformational changes can range from very local (e.g.,
> unwinding of a helix) to very diffuse (e.g., subtle but significant rigid body shifts between two domains.) In the first case,
> it would probably make sense to do a global least-squares fitting, but in the latter, one would do better to fix one of the
> domains, and show the shift in the other domain. These cases, however, presuppose that one knows which type of case
> one is dealing with. This could be done by guesswork and trial-and-error, but does anybody know of an approach (e.g., a
> program) to define the most reasonable way to think about a given conformational change? Variable-size sliding-window
> least-squares superpositions with comparisons of local versus global rmsd's come to mind, but I do not know whether
> this has been implemented anywhere, and would not know readily how to set the parameters thereof either.
DynDom may do this, but I'm not familiar with the program. (It's in CCP4 now, I think)
If you're just trying to get a reasonable superposition and don't care very much about the resulting statistics, you can usually use a much simpler method called a "sieve-fit", described in these references:
In practice, the procedure described in the second paper generally worked very well for the intended purpose of visualizing any arbitrary conformational change in the PDB clearly. The code that actually performs this isn't distributed as far as I know; however, it should be relatively trivial to re-implement using CCTBX or something equivalent.
PyMOL's "align" command also does some kind of iterative optimization by throwing away outliers, but it's much less aggressive and appears to try for the best global fit, excluding loops etc.
-Nat
