Phase Improvement and Interpretation Manual |
previous |
Chapter 2 |
|
Phase Improvement and
Interpretation Manual - Control
Copyright |
© 2001-2006 by Global Phasing Limited |
|
|
All rights reserved. |
|
|
This software is proprietary to and embodies the confidential
technology of Global Phasing Limited (GPhL). Possession, use,
duplication or dissemination of the software is authorised only
pursuant to a valid written licence from GPhL.
|
|
Documentation |
(2001-2006) Clemens Vonrhein |
|
Contact |
sharp-develop@GlobalPhasing.com |
Contents
The "Phase Improvement and Interpretation Control Panel" lets
you do most of the subsequent steps (after running SHARP for heavy atom refinement and phasing)
necessary during a structure solution for improving the initial phases
and getting the best possible electron density map. However, some tools
are in a more advanced stage of development than others. Especially,
everything concerned with non-crystallographic symmetry (detection and
use) is still rudimentary. If you have any comments or remarks please
get in contact with us at sharp-develop@globalphasing.com.
Since several similar procedures might be run (within the same
sub-directory in your sharpfiles/logfiles directory) we need to
distinguish these from each other. This is done through the solvent
fraction XX.Xpc (expressed as a percentage
with only one digit being used). The interface will warn you if you try
to run a similar procedure with the same solvent fraction.
This section contains all options specific for solvent flattening/flipping.
This is the solvent
content of your crystal. It has to be unique for this
procedure: just change it in steps of 0.001. Although the solvent
fraction will slightly differ it won't have too much of an influence.
This doesn't have to be the correct solvent content of your crystal: it
is the one used for calculating the solvent envelope.
The resolution range to use throughout the procedure. Make sure to take
the phase quality into
account.
If using DM at any stage of the solvent flattening procedure(s), this
solvent content will be used for data scaling (see DM documentation). It should always be the real
solvent content of your crystal. The default of "0"
will make this the same as the solvent
fraction.
If this differs from the default of "0", the first ten cycles
of solvent flattening will change the solvent content used for defining
the solvent envelope from this value to the solvent fraction in 10 equal steps. This
could be helpful if the starting phases are of particular bad quality
and a gradually increasing solvent content at the beginning of the
solvent flattening will make sure, that no potential protein regions are
flattened (and therefore never recovered).
SOLOMON defines the solvent envelope by
using a local standard deviation map. A sphere at each grid point is
used to calculate the rmsd of electron density within this sphere. The
protocol for solvent flattening/flipping used here allows this sphere to
slowly decrease during the various cycles. Tests suggest that a good
starting value is the resolution where the figure-of-merit drops
below 0.5. The final radius is usually the high resolution limit used.
The protocol for solvent flipping using SOLOMON needs a flipping
factor. The default of "0" will make sure, that the
correct (gamma correction) value is used.
During solvent flipping using SOLOMON, the
density in the solvent and protein region are adjusted to have a ratio
defined by the mean density in the solvent and in the protein
region. Water should have a mean density of 0.32
e/Å3. For proteins this should be about 0.43
e/Å3, DNA should have about 0.60
e/Å3. For mixtures of protein and DNA/RNA you might
need to adjust this.
The solvent flattening/flipping procedure used will truncate the lowest
and highest grid points in the "protein" region. The defaults
(lowest 40 % and highest 1 %) seem to work in most cases. If you suspect
large peaks (e.g. caused by heavy atoms) in the protein region you might
want to increase the fraction of highest peaks truncated. See also the
documentation for SOLOMON.
In some cases - especially when only low resolution data or phases are
available - it might be beneficial to start with a simple run of DM (solvent flattening and histogram matching)
before switching to solvent flipping with SOLOMON. This can lead to a better starting map
for the iterative solvent flipping procedure.
If some kind of model is available (partial molecular replacement
solution, partially built model, automatically built model etc) it can
be included in the solvent flipping procedure. This can greatly improve
the starting cycles of density modification by providing a much better
initial solvent envelope. If some loops or flexible parts of your model
are missing in the solvent flattened map it might be because these
regions were initially included in the solvent part of the envelope and
therefore flattened.
Note: make sure that the experimental phases (ie heavy atom
sites) and the model are on the same origin and have the same
handedness!
Currently, only closed local symmetries of the form Cn can be
located. More complicated forms of NCS have to be dealt with externally.
During NCS detection using GETAX, this is
used to estimate the volume of your "protein" monomer,
ie the part of your structure obeying the NCS.
During NCS detection, for each rotation determined from a self-rotation
function, a certain number of possible translation solutions can be
tested. Your own set of peaks in a self-rotation can be supplied by
editing a file selfrot.lis and
adding polar angles omega, phi, kappa (for angle convention see POLARRFN documentation). Each of these rotations
is fed into GETAX to find possible centres
of a Cn rotation axis.
Although the assumption for NCS detection using GETAX is, that a pure Cn rotation is
present, a slight deviation of this ideal scenario might occur. This
might not have a great effect in finding the (roughly) correct position
of the rotation axis. However, during refinement of the operators, the
use of a pure multimer mask when a monomer mask is required can drown
the correct solution in noise. Therefore, a set of possible monomer
masks is tested, each kappa/N angles apart (where kappa is the rotation
angle around the rotation axis - 180, 120, 90 etc - and N is the number
of monomer positions to be tested).
This options is probably only useful, when the correlation map from
GETAX shows clear stretches of high
correlation (corresponding to a possible rotation axis position) but the
following NCS operator refinement using a multimer mask is unsuccessful.
In some cases, local rotation axes can be parallel to crystallographic
rotation axes. These would be buried under the large crystal
symmetry peaks in the self-rotation. In these circumstances, this option
could be turned off.
In some cases, these parallel local symmetry axes are visible in a
native Patterson function.
The program package ARP/wARP is used for
automatic building and tracing in a solvent flattened map. If you want
to use this feature, you have to have the latest version installed and
configured (ask your SHARP site
administrator about that).
We use a script that is more or less identical to the distributed arp_warp.sh script: it differs in that it
only supports the warpNtrace protocol and has a different error
handling. So you should be able to use the warp.par file produced through this
interface with your own ARP/wARP
installation without any further changes.
The total number of protein residues that ARP/wARP should be building. This is not the number of residues per chain/monomer, but
the total number of protein residues in the asymmetric unit.
Two parameters control the number of cycles that ARP/wARP should be running in automatic building
and tracing mode (warpNtrace). This defines the overall number of
cycles after which it will stop. The default of 100 should be a good
starting point - but see the ARP/wARP
documentation for more details.
The number of cycles, after which ARP/wARP
will try to re-interpret the density/model. At this stage, automatic
building and (possibly) side-chain docking is done. In most cases, the
default of 10 should be adequate - but see the ARP/wARP documentation for more details.
If a sequence file is available, the pull-down menu should allow for its
selection. ARP/wARP will try side-chain
docking at various stages of the warpNtrace procedure. See the
ARP/wARP documentation for more details.
The auto-building will always use experimental phases during
refinement. These can come either direct from SHARP or from the
solvent flattening procedure. In the latter case the
Hendrickson-Lattmann coefficients from the last DM run are used - and
these should be automatically adjusted so that at later stages of the
automatic tracing/refinement they are appropriately dampened.
The fraction of successfully traced main chain can be used to dampen
the contribution of experimental phases during each refinement
cycle. If phase information after
solvent flattening is used this is recommended (the phase
probabilities after density modification are nearly always
overestimated). If you prefer to use your SHARP phases directly it
should not be necessary - since SHARP outputs a reliable estimation of
phase probabilities.
The weight between X-ray and geometry term in the refinement can
automatically be adjusted to give a reasonable geometry of the final
model. The rms deviation on bond distances is used for this.
To speed up the process of refinement, the free R-factor can be used
to stop the refinement cycles between rebuilding cycles. A certain
degree of fluctuation is allowed, but too large increases in Rfree
will trigger the refinement cycles to be abandoned and the switch to a
building cycle.
Some tools are provided that can help you using or analysing the results
from the various phase improvement and interpretation protocols.
A simple procedure can help you in finding the solvent content that
gives the best overall density for a given
protocol. If you ran three different solvent flattening runs
where only the solvent content was changed, a parabolic fit to these
three values should give an idea of the optimal solvent content. This is
exactly the strategy used by the "Solvent-flattening
(optimising)" protocol.
Note: this will obviously not work if any of these runs produced
complete nonsense. See the remarks in each solvent flattening log-file
for possible problems.
This tool gives the possibility to extract Hendrickson-Lattmann
coefficients from
- solvent flattening,
- automatic building or
- a model
These can then be used as external phase
information in SHARP to help refinement
of heavy atom parameters.
In case you have various versions of SHARP installed and you want to
use a version of SHARP later than 2.0.0 you should make sure that
these are written into a combined file.
To calculate ordinary difference Fourier maps (isomorphous/dispersive
and anomalous differences) any of the solvent flattened phases can be
used. All amplitudes and anomalous differences in the REFL01.mtz file (ie the data file used for
the corresponding SHARP run) will be used.
Although it is recommended to use the residual (log-likelihood
gradient) maps, this is an easy way to check existing sites, find
sites in new, unused datasets etc.
Most of the output produced during any of the above steps should be
self-explanatory. However, here are some basic explanations about the
various results you can obtain.
The density modification can be run either for each solvent content by
hand or using the 'self-optimising' option. In any case, each run with
a specific solvent content will produce a log file with a lot of
information. This will be explained here.
This value compares the standard deviation of electron density in the
solvent region and the protein. Since we expect relatively flat
solvent (low sd) and lot of features in the protein (high sd), this
value should steadily go down during density modification. Obviously,
the determination of where solvent and protein is is crucial. So
again, the very first solvent mask determination has a large effect on
this quality indicator.
If a newer version of SHARP/autoSHARP is used (post 2.0.0) an additional step of
scaling using ICOEFL is performed. The final R-factor is given. This should
be similar to the ones coming from SFALL and/or RSTATS. Obviously, it should go
down during the iterative density modification - but be aware of the
bias problem intrinsic in this type of density modification!
This is a simple R-factor between the structure factor amplitudes from
the modified map and the 'observed' data. This should be similar to
the ones coming from ICOEFL
and/or RSTATS. Obviously, it
should go down during the iterative density modification - bu be aware
of the bias problem intrinsic in this type of density modification!
This is a simple R-factor between the structure factor amplitudes from
the modified map and the 'observed' data. This additional scaling is
done right after the calculation of structure factors (in SFALL). The
value should be similar to the ones coming from ICOEFL and/or SFALL. Obviously, it should go
down during the iterative density modification - bu be aware of the
bias problem intrinsic in this type of density modification!
During calculation of appropriate weights for the modified structure
factors (and finally phase combination with the experimental phase
information from SHARP), the correlation coefficient on E**2 values is
calculated in SIGMAA (between the structure factor amplitudes of the
'observed' data and the modified map). This is a good indicator of the
quality of the map - at least at the very first cycle (when the
procedure has still no bias introduced). In our experience, a value of
lower than 0.1 at the very first cycle usually points to a very bad
starting map (and therefore bad starting phase values and/or data).
Here, only the overall value is
given. But it might be a good idea to look at these values as a
function of resolution (just follow the link 'directory listing' to
get to the complete log files). The logfile for the first cycle is
also kept.
A good indication if the starting map is of promising quality, is to
compare this value at cycle 1 for each of the two possible hands (if a
change in handedness is possible). In good cases, you should see a
significant difference in this value between the two hands.
At the end of the solvent flipping procedure (and in fact, after the
last cycle of solvent flattening), an additional density modification
calculation using DM is
performed. This will result in another set of phases as well as a set
of Hendrickson-Lattmann coefficients. The latter can be fed back into
SHARP as external phase information to help the refinement of the
heavy atom model as well as the calculation of residual maps.
Last modified: Fri Sep 15 14:58:02 BST 2017