Ian is right and, using the same logic, the most appropriate solution
appears to be to use the Sigmaa estimates from the last refmac run
(based on the free reflections). I thought there was a way to read these
in from the MTZ but can't find it in the manual. If I understand it
correctly, Sigmaa estimation for ML refinement is controlled through the
SCALE keyword, which is rather confusing (see manual excerpt below).
The only option I see is to set SCALE MLSC WORK to let sigmaa be
estimated from the working set reflections. Not perfect, but at the end
of the refinement sigmaa values are close to 1 for most of the
resolution range so it probably doesn't matter much in practise. If
anyone knows how to get sigmaa estimates from the free reflections in
the first cycle, or read them in from an MTZ, and then keep those sigmaa
values fixed while refining the model against all data then please let
us know.
Bart
=======================================================================
REFMAC manual excerpt
SCALe MLSC [ <NCYC <ncyc> | WORK | FIXBulk SCBUlk <scbulk> BBULk <bbulk> ]
For example:
SCALe MLSC FIXBulk BVALue 100.0 SCVAlue -0.1
The SigmaA estimate is generally fitted to the normalised Free
reflections using a 4 parameter equation of an analogous form to the
bulk scaling:
SA = SA0*exp(-T0*s^2) * (1- SA1*exp(-T1*s^2))
This keyword controls the estimation of SigmaA. Subkeywords:
FIXBulk
The option FIXBulk to fix parameters can be evoked in the same way
as for the SCAL LSSC options, but should only be used with care!
NCYCle <ncyc>
[Default <ncyc> = 10]
Use <ncyc> cycles to determine the parameters.
WORK
[Default Sigmaa is calculated against the FREE set of reflections]
The keyword WORK changes this to determine the scale from the
WORKing set of reflections.
Ian Tickle wrote:
I believe the recommended procedure is to refine against all data before
submission, using the same set of parameters and weights as for the last
refinement against the working set alone, after all why throw away 5%
(or even 10% in some cases) of your hard-won data? The purpose of Rfree
is 1) to check for gross errors in the model, i.e. Rfree should not be
significantly greater than (or indeed less than) its expected value
(which depends essentially on the observation/parameter ratio), and 2)
to check for over-parameterisation and/or incorrect relative weighting
of X-ray terms & geometry (i.e. Rfree should be minimal at convergence
of refinement at maximal overall likelihood for an appropriate choice of
variable parameters and weights). Once your structure has passed these
tests satisfactorily there's no further need for the test set.
Basically you do all your cross-validation before you abandon your test
set.
There is however a difficulty with this procedure, and it may not be
practical to carry it out: namely ML refinement uses the test set to get
unbiased estimates of the sigma-A's, so it may not actually be formally
correct to refine without a test set using e.g. Refmac. Other
well-known refinement packages (e.g. Shel-X) don't have this problem.
-- Ian
-----Original Message-----
From: CCP4 bulletin board [mailto:[EMAIL PROTECTED] On
Behalf Of David Briggs
Sent: 30 January 2007 16:10
To: [email protected]
Subject: Re: [ccp4bb] problem with anisotropic refinement
using refmac - Sacrosanct R-free
"...They might
also get rectified when you use all of your data for
refinement, meaning
when you are not using any R free..."
Sorry to go off on a little tangent here - but doesn't
refining against ALL of your data generally mean that your
RFree is no longer your "free" and cannot be used for
cross-validation...
I was always taught that the reflections within the test set
are sacrosanct - is this not also the case for circa atomic
resolution data? Perhaps someone more learned than I could
shed a little light on this...
Dave
Hallo to all out there,
>
> I have a problem refining a structure against a 1.33
A resolution data
> set. Using REFMAC (version 5.2.0019, also tested
version 5.3) with
> isotropic B-factor refinement resulted in an valid
model with R /
> R(free) of 17.8 / 19.2%.
>
> To finish the model I tried to use anisotropic
refinement which should
> be possible/reasonable because of an observable to
parameter ratio of
> about 2.8. This refinement yielded to a much better R
/ R(free)-Factor
> of 14.1 / 16.8% and converged after about 15 cycles
of anisotropic
> refinement. However the problem is that running a few
cycles of
> refinement resulted in warnings, logged as "Problem
in MAKE_U_POSITIVE".
>
> So what would be the right solution to overcome this problem?
>
> Some further information about the data set:
>
> spacegroup P 6(5) 2 2, one protomer per ASU
> cell axis 76.615 76.615 209.787
90.00 90.00 120.00
> unique reflections 83156 (6220)
> Completeness 98.6 % (87.9%)
> I/Sigma 21.4 (3.5)
> Rmrgd-F 5.9% (35.0%)
>
> Maybe the refmac-script will be of some help (some
other BFAC restraints
> and SPHE/RBON parameter tested, the following example
takes care of
> reasonable distribution of anisotropy):
>
> #!/bin/bash
> refmac5 hklin ../gz_ccp4.mtz \
> hklout gz_aniso_01f.mtz \
> xyzin ./gz_iso.pdb \
> xyzout gz_aniso_01f.pdb \
> libin ../llp_citrat_fitted.cif \
> << end_ip > refmac.log
> LABI FP=F_cit_01 SIGFP=SIGF_cit_01 FREE=FreeR_flag
> LABO FC=FC FWT=FWT PHIC=PHIC PHWT=PHWT DELFWT=DELFWT
PHDELWT=PHDELWT
> FOM=FOM
> NCYC 20
> REFI TYPE RESTRAINED
> REFI RESI MLKF
> REFI METH CGMAT
> REFI RESO 25 1.33
> REFI BREF ANISOTROPIC
> SCAL TYPE BULK
> SCAL LSSC ANISO NCYCLES 10
> SCAL MLSC NCYCLES 10
> WEIG MATRIX 1.25
> SPHE 30.0
> RBON 30.0
> BFAC 0.5 2.0 4.0 4.0 6.0
> MAKE CHECK ALL
> MAKE HYDROGEN ALL
> MAKE HOUT NO
> MAKE PEPTIDE NO
> MAKE CISPEPTIDE NO
> MAKE SSBRIDGE NO
> MAKE CHAIN YES
> MAKE SYMMETRY YES
> MONI MANY TORS 10 DIST 10 ANGL 10 VAND 10 PLANE 10
CHIR 10 BFAC 10 BSPH
> 10 RBOND 10
> BINS 20
> PNAM gz
> DNAM gz
> USEC
> END
> end_ip
>
> The final refinement statistic:
>
> Resolution limits = 25.000 1.330
> Number of used reflections = 81889
> Percentage observed = 98.6122
> Percentage of free reflections = 1.5000
> Overall R factor = 0.1409
> Free R factor = 0.1681
> Overall weighted R factor = 0.1348
> Free weighted R factor = 0.1641
> Overall correlation coefficient = 0.9763
> Free correlation coefficient = 0.9688
> Overall figure of merit = 0.9183
> ML based su of positional parameters = 0.0274
> ML based su of thermal parameters = 1.5420
> rmsBOND = 0.014
> rmsANGLE = 1.569
>
> Thanks in advance,
> georg zocher
>
>
--
---------------------------------------
David Briggs, PhD.
Father & Crystallographer
www.dbriggs.talktalk.net
iChat AIM ID: DBassophile
Disclaimer
This communication is confidential and may contain privileged information intended solely for the named addressee(s). It may not be used or disclosed except for the purpose for which it has been sent. If you are not the intended recipient you must not review, use, disclose, copy, distribute or take any action in reliance upon it. If you have received this communication in error, please notify Astex Therapeutics Ltd by emailing [EMAIL PROTECTED] and destroy all copies of the message and any attached documents.
Astex Therapeutics Ltd monitors, controls and protects all its messaging
traffic in compliance with its corporate email policy. The Company accepts no
liability or responsibility for any onward transmission or use of emails and
attachments having left the Astex Therapeutics domain. Unless expressly
stated, opinions in this message are those of the individual sender and not of
Astex Therapeutics Ltd. The recipient should check this email and any
attachments for the presence of computer viruses. Astex Therapeutics Ltd
accepts no liability for damage caused by any virus transmitted by this email.
E-mail is susceptible to data corruption, interception, unauthorized amendment,
and tampering, Astex Therapeutics Ltd only send and receive e-mails on the
basis that the Company is not liable for any such alteration or any
consequences thereof.
--
==============================================================================
Bart Hazes (Assistant Professor)
Dept. of Medical Microbiology & Immunology
University of Alberta
1-15 Medical Sciences Building
Edmonton, Alberta
Canada, T6G 2H7
phone: 1-780-492-0042
fax: 1-780-492-7521
==============================================================================
