In article <4oaa9b$dpq at madeline.INS.CWRU.Edu>, hi712 at cleveland.Freenet.Edu (Ganesh Jawahar Swaminathan) writes:
|>|> I am trying to solve the structure of a protein with MR using a
|> model that has low homology to the protein I am trying to solve
|> (45%). Rotation and Translation Function Results are quite encouraging
|> but refinement in X-PLOR show the crystallographic R to continually
|> decrease while the Free R stays at a high level. I end up with
|> situations of 17%R and 46%Free R.
|> I guess this problems may be due to low sequene homology. But what is
|> the real significance of this value other than overfitting the data
|> to the model ? Also what are the ways out of this problems.
my first MR solution had 25% identity to the target, so 45% actually
sounds quite good to me - assuming that is %identity and not some homology
matrix score (you should specify). the main suggestion I would make is GO
SLOWLY. Start with rigid body refinement only. make whatever substitutions
and changes you can to fit the new molecule at every step. move from there
to minimization only. simulated annealing has a larger "radius of convergence",
but it also has more potential for model bias if you are over-fitting, and it
sounds like you are. And put a low weight on the X-ray term as you refine in
the early stages. you can always make the model fit the map, but if you are
doing it at the expense of geometry or free-R, you are actually hurting
yourself. I typically ran the CHECK.INP job and then used a value ~ 0.2 times
this.
if you can get any independent experimental phases into the picture it can also
help reduce model bias. if you have any heavy atom data sitting around, try to
solve it using the MR solution to phase 'difference Fouriers'. I was able to
solve a single derivative, then combine that with the MR phases and extend from
there. helped a lot.
remember, your job is to solve the structure, not to bring the R down
(that will take care of itself when you get the model right); so if
it takes you a few weeks or even months of fitting and refining to drop R
into the low 30's, don't worry about it. you have to get away from the belief
that you can take the MR solution, run one simulated annealing job
and you're done.
=======================================================================
They saw the staggering jewels of the night
in their infinite dust and their minds sang with fear.
- Douglas Adams
=======================================================================
Dave Schuller
University of California-Irvine
schuller at uci.edu
modern man in a post-modern world
