Yes, agree overall.
What I plan to implement in phenix.refine and then play with at some
point is that loop I wrote before:
for cycle in cycles (until convergence):
- do real space refinement (minimize T = Exray*weight +Egeom w.r.t.
model params);
- do reciprocal space refinement (minimize T = SUM(rho_obs -
rho_calc)^2*weight +Egeom w.r.t. model params)
(- do other things, like we do now: model bulk and ordered solvent,
etc...)
Based on the literature review, the overall feeling is that this should
work better than either one separately: real or reciprocal space
refinement.
Pavel.
Anastassis Perrakis wrote:
On 10 Aug 2007, at 20:12, Pavel Afonine wrote:
Anastassis Perrakis wrote:
On 10 Aug 2007, at 18:59, Pavel Afonine wrote:
Hi Mike,
the best is to do both in a loop:
for cycle in cycles:
- do real space refinement;
- do reciprocal space refinement
Well - thats what we all do - right ?
The real space refinement can be done either with the tools from
Chapman at al,
or interactively by the user or by (here I go again ... ) ARP/wARP
or Resolve etc etc.
In all cases we fit the model to the map ... only the tools differ.
I was not clear enough, sorry. By refinement I meant the
mini(maxi)mizer-driven optimization of a goal function w.r.t. model
parameters, and not atoms re-shuffling manually (looking at map) or
using a program.
Clear clarification ... but there is hardly a difference I would say.
What ARP/wARP, or Coot, or O, or I am sure Resolve, do at some point
is exactly to use a mini(maxi)mizer-driven optimization of a goal
function w.r.t. model parameters to fit the model to the real space
density better.
ARP/wARP uses a simplex minimizer with a real space correlation
function and a torsional parameterization, and if I remember Michael
Chapman's papers
he uses a powell minimizer in a simulated annealing protocol and a
cartesian parameterization, and if I again recall Paul uses the BDGFS
minimizer in cartesian space, but they all do the same things at one
point or another, which what you describe. If before that one used the
mouse or a clever algorithm to build a model, this is always refined
as you describe in real space, before entering 'real' refinement.
Tassos
Pavel.