On Sat, Oct 05, 2019, Carlos Simmerling wrote:
>Is anyone actively developing code to do cryo-em refinement or structure
>fitting with Amber?
I'm working (some) on this, hoping to improve the target functions
("emap") that Xiongwu put into sander and pmemd. My idea is to use the
Fourier representation of the map, with the target function being the
squared magnitude of the vector difference between Fobs and Fcalc,
rather than just the difference in amplitudes (as is done in
crystallographic refinment). Hence, the phases in the cryoEM map would
be targeted as well amplitudes.
This is going to take advantage of GPU reciprocal space refinement codes
that Dave Cerutti and I have been working on. It's not clear (yet) how
this proposed target function will differ from the real-space functions
often used -- I've been hoping to try something a bit new, and not
(just) port into Amber what CHARMM/NAMD/Gromacs are already doing.
[Porting emap itself, or a similar correlation-based target function, to
GPUs would also be an attractive option: we could leverage good force
fields plus replica-exchange or other enhancments in sampling to provide
some pretty powerful tools to users.]
I'm just getting started, but have some pointers to what should be
some good test cases. I'd be happy to compare notes with other
approaches.
....dac
p.s.: I do think that anyone wanting to enter this area should take
a careful look at the ISOLDE program from Tristan Croll. This is a
combination of MD-guided refinemnt (built on openMM) with interactive
modeling. Tristan makes this go fast by just refining parts of the system
at a time, but the provision for visual feedback makes this approach very
attractive.
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat Oct 05 2019 - 19:30:02 PDT
