*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
Yuhe Liang wrote:
I am refining a 1.4A resolution structre with CNS.
The model is now almost ok and the Rfree is around
0.23. The problem now is the atom coordinates of
the model don't change when I do minimize job, only
a overall B-factor refinement was done. And the
minimize procedure terminated after 5 steps when
Rfactor didn't change. I tried to change the resolution
limit and wa setup, but it doesn't work. Simulated
annealing can change the coordinates, but after that,
minimizing can't refine more on the results.
I also tried to use a not-well-refined model as the
input model, then the minimizing had effect!?
Did you have a message in the log file like:
"Line search terminated
LBFGS: maximum number of function evaluations reached"
I had a similar problem (mionimization terminates early,
output coordinates identical to input) when the geometry
of a prticular region was very bad. Doesn't sound likely
since this is your well-refined model, but who knows?
Here is what I wrote at the time. Look also at Harry Greenblatt's
problem and see if your well-refined model has some
coordinates = 9999.000
Ed
-------------------------------------------
Edward A. Berry wrote:
I ran into a problem as reported by Harry Greenblatt and Sacha Uljon,
In which minimize.inp quit early, writing out unchanged coordinates,
after
"Line search terminated
LBFGS: maximum number of function evaluations reached"
I was also not building the unknown atoms like Harry, but
unlike him I was deleting the unknown atoms so there
was nothing at 9999.999.
Taking a clue from Harry's solution I deleted parts of the structure-
first prosthetics and ligands, then each chain one at a time.
If one chain was deleted the problem went away,
this further narrowed down to a stretch of 3 residues
that could be deleted to give normal operation.
Taking a look with O the geometry was very bad- specifically
O failed to connect the peptides at the ends of the stretch.
I did a few cycles of refi-zone over the strectch in O, wrote
the coord out, and refinement worked again.
I guess if there is a real problem with something in the model,
the energy terms get screwed up and minimize cannot calculate any shifts.
Try to localize the problem and examine or delete.
Harry M. Greenblatt wrote:
BS"D
Dear All,
Some weeks ago, I reported trouble with going from strict NCS to
restrained NCS (cns_solve 1.0). The program terminated prematurely,
complaining of a non unitary matrix. A few people offered suggestions,
but none gave the answer. Gerard's answer, however, was the one that
got me on the trail of the solution. He told me to make sure that each
monomer of the generated explicit dimer was identical to the other.
They were, in fact, but the search revealed a different problem, which I
guess is a bug.
As a first step I removed the bound inhibitor, waters, and
carbohydrates, to give a plain protein case. The program no longer
complained about the matrix, but the refinement (minimize.inp) blew up
quickly and stopped with the complaint "...line search terminated...".
I had checked are rechecked the co-ordinates on the graphics, etc., and
there was no problem with the NCS dimer. But the program kept reporting
the rms difference between the two NCS groups to be ~32A.
I had successfully run restrained NCS on this crystal system before,
and so I went back to look there. The difference between the cases was
that previously I had allowed generate.inp to create real co-ordinates
for my missing atoms. In the current case, I did not build these atoms,
and so they all had x,y,z co-ordinates of 9999.000. When I changed this
option in generate.inp, and got real atomic positions in the current
case, minimize was happy.
So there seems to a bug in the way cns_solve handles dummy atoms in
restrained NCS. Anyone with similar experiences?
HMG
