[ccp4bb]: exploding atoms during final refinement

[Jan Abendroth <Jan.Abendroth@uni-koeln.de>]

Dear all,
some strange things happen during the last cycles of my refinement ...

The situation:
I am refining quite a large structure consisting of six 50kDa Monomers
(6*458=2748 residues + some 3800 water + some buffer) with a 1.3 AA
data-set and refmac5. I always use a script like the one attached 
incuding contributions of hydrogen (without refining them). After
introduction of anisotropic B-factors and some few multiple
conformations the refinement is at is final stage with R-values around
R-work=14.5% and R-free 17.5%. When I changed the restrains on the shape
the displacement ellipsoids (SPHERE) from default 2 to 5 or higher both
R-values dropped by 1% but some severe problems occurred and the
refinement diverged:

The problems:
>From one cycle to the next within refmac the B-factors of one atom in
one of the last residues of the sixth monomer suddenly grew a lot (>100
and much more in following cycles) and became very oblate. The R-values
inceased by between 1 and more than 5% -  the refinement diverged rather
than converged. Once a monomer was affected more and more atoms in the
same monomer showed these strange B-factors in the following cycles.
This comes along with an increasing number of "Eigenvalue"-,
sphericity-outlier- and rigid-bond-outlier-warnings

Trials how to fix it:
I have tried the following things in order to make the refinement
convergent again:
- inclusion of ncs: code 4 did not help, code 6 caused a core dump
- fixing Babinets Solvent B to 200: didn't help
- restraining B-shifts (HOLD BFAC from def=3 to 10): didn't help
- newer refmac-5-version (5.0.28): things got much worse!

- starting with isotropic B-factors: helps for a while (5 big cycles)
- resetting SPHERE from 5 to 2: B-factors remain stable, but both
R-factors get 1% higher again

Questions:
- Why does this strange behaviour start in a clearly defined region of
the structure?
- Why does this stay in one monomer, the others have the same
sequece/structure?
- Why are there severe differences between the refmac-versions?
- Does refmac have some problems with large structures?
- Does it make sense to return to tighter SPHERE restraints in spite of
1% larger R-values?
- Does anyone have some similar expriences or some experiences with the
SPHERE restraints?

Thanks a lot for any hints and suggestions!
Best regards

Jan

-- 
------------------------------
Jan Abendroth
Institut fuer Biochemie
Universitaet Koeln
Zuelpicher Strasse 47
D-50674 Koeln

Tel: +49-(0)221 470 6455
Fax: +49-(0)221 470 5092
#!/bin/tcsh -ef
###################################################################
#    Generate hudrogens and calculate contribution of them.
###################################################################

set last = 0
set count = 0
set cycles = 5	# grosse Zyklen
set runden = 5	# refmac Zyklen

while ($count != $cycles)

echo '*******************************************************************'
echo  $count
echo '*******************************************************************'
@ curr = $last + 1


nice -20 \
hgen \
	XYZIN  refmac-aniso-h_${last}.pdb \
	XYZOUT hydrogen_${curr}.pdb \
	<< eof | tee hgen_${curr}.log
HYDROgens SEPArate  ! Hydrogens in seperate file
eof
#
###################################################################
#     Calculate contribution from hydrogens. NOSCALE option should
#     be used to preserve absolute scale of structure factors of
#     hydrogens
###################################################################
nice -20 \
sfall \
	XYZIN  hydrogen_${curr}.pdb \
	HKLIN  ./../daten/dhyd_1295-unique.mtz \
	HKLOUT hydrogen_${curr}.mtz \
 	<< eof | tee sfall_${curr}.log
NOSCALE
MODE SFCALC XYZIN HKLIN
LABI FP=F_NAT  SIGFP=SIGF_NAT  FREE=FreeR_flag
LABO FC=FC_hyd PHIC=PHIC_hyd
!Refinement parameters
RESO 50 1.295
BINS 20
END
eof
#
##################################################################
#    Refinement with refmac, inclusion of hydrogen contributions 
#    as FPART1, PHIP1
##################################################################    
nice -20 \
/usr/local/refmac_5.0.28/refmac5 \
	xyzin  refmac-aniso-h_${last}.pdb \
	hklin  hydrogen_${curr}.mtz \
	xyzout refmac-aniso-h_${curr}.pdb  \
	hklout refmac-aniso-h_${curr}.mtz  \
	<< eof-refmac | tee refmac_${curr}.log
TITLE refmac5 with hydrogen-contributions
LABIN FP=F_NAT SIGFP=SIGF_NAT FPART1=FC_hyd PHIP1=PHIC_hyd FREE=FreeR_flag
LABO  FC=FC PHIC=PHIC FWT=2FOFCWT PHWT=PH2FOFCWT DELFWT=FOFCWT PHDELWT=PHFOFCWT
MAKE_RESTRAINTS HYDR NO CISP YES
REFI TYPE REST RESO  50 1.295
REFI RESI MLKF 
REFI BREF ANISO
REFI METH CGMAT
WEIGgt MATRix 5
SPHEricity 5
SCAL TYPE BULK
NCYC $runden 
eof-refmac

if ($status) exit
#
@ last++
@ count++
end