On Mon, Oct 07, 2019, Xiongwu Wu wrote:
>I think your target function would be advantageous for structure fitting,
>because translational conformational search can handled more efficiently
>with FFT. How about orientational space search? Does every orientation
>change need recalculate FFT? High noisy nature of EM maps also need to be
>taken into account.
I had not thought about using my new ideas for (rigid) fitting. The
codes you have do (for my few examples) a good job with that. They are
rather slow, but (a) I usually only do that once; and (b) the rigid
fitting has often already been done by my collaborators, so I have a
starting point that is already fit to the map. Hence, I've never
thought much about trying to optimize translational or orientational
fits.
Rather, I'm thinking of this for MD-based refinement. The "emap" target
function has the strength (and limitation) that it tries to put every
atom into density. This does not have any (direct) way to provide a
bias against having density with no atoms in it, and relies on
non-bonded contacts to keep more than one atom from occupying the same
map density.
I don't think there is any "right" approach for all problems. As you
point out, some maps can be very noisy, and many problems have missing
parts of the model (so that one, indeed, expects to find density with
no corresponding atoms in it). So, I'm initially probably going to
focus on "cleaner" problems, where forcing the map fourier coefficients
from the model to match those from the experiment seems likely to
be a good target.
All this is really new for me, so I'd appreciate both suggestions for
what to do, and for useful test cases where one could compare different
approaches.
...thx...dac
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon Oct 07 2019 - 09:00:02 PDT
