The “Automated structure solution - Crank2” experimental phasing task
builds a model automatically from intensities or amplitudes from
single or multiple-wavelength anomalous diffraction (SAD/MAD) or
single isomorphous replacement (SIRAS) experiments. The user can
optionally choose to input a partial model and start or end the
automated pipeline at each of the steps below:
Substructure detection with
SHELXD
or PRASA (to be released in 2016) using FA values obtained from
SHELXC,
AFRO or
ECALC
Substructure improvement & phasing with
REFMAC,
SHELXE,
or BP3
Hand determination with
SOLOMON and MAPRO or
SHELXE
Density modification with SHELXE or
PARROT or SOLOMON
with phase combination with REFMAC or MULTICOMB
Model building with
Buccaneer,
ARP/wARP
or SHELXE iterated with model refinement with REFMAC
A pipeline can be started or stopped at any stage (1.1) For
example, if an anomalously scattering substructure has been
determined already, it can be input in PDB format by selecting to
start the pipeline at “Substructure improvement” (SAD experiment) or
“Substructure phasing” (SIRAS,MAD).
A partial protein model can be input in PDB format (1.2), to be
rebuilt using the SAD data. If the partial model is biased, the
option to exclude the non-anomalous atoms from the partial model can
give better results. It is recommended to try both approaches.
Inputting a protein sequence (1.3) is beneficial for model
building. If it is not available, click off the Input protein
sequence option and input the number of protein residues per monomer.
Input the substructure atom element symbol (i.e. Se for Selenium) and
the number of substructure atoms in the asymmetric unit (1.4).
The number of substructure atoms in the asymmetric unit is the number
of substructure atoms in your protein multiplied by the number of
protein molecules that are present in the asymmetric unit. For
sulfur-SAD phasing of structures containing disulfide bonds, the
sulfur atoms that make such bonds should be counted separately in the
number of substructure sites, and the minimum distance between sites
should be given as 1.5A in the “Advanced options” section (3.3).
If the anomalous resolution is worse than about 2.1A, the disulfides
will not be resolved, so the number of disulfides (i.e. “S-S pairs”)
should be specified and the minimum distance increased to 3.5A, but
the number of sites is still the total number of anomalously
scattering atoms.
Input the anomalous reflections data in a merged MTZ file containing
the Bijvoet/anomalous intensities or structure factor amplitudes
(1.6). Alternatively, unmerged or merged XDS (XDS_ASCII.HKL),
HKL2000 (.sca) or SHELX (.hkl) files can be input (1.5).
Inputting unmerged files is recommended if SHELXC is used since it
can make use of it for improved estimates of FA values and reporting
of statistics such as the anomalous correlation coefficient
(CCanom1/2). If non-MTZ reflection data is input, fields for crystal
cell dimensions and spacegroup appear and are automatically filled by
the ccp4i2 interface if the information exists in the input file. If
they have not been filled in, please input them.
Finally, the Crank2 pipeline requires the anomalous scattering
coefficients f’ and f” for the substructure atom at the wavelength of
the data collected. The interface can automatically fill in the
theoretical values for the scattering factors but it is very
important to check these, especially for wavelengths close to the
absorption edge. The values from a fluorescent scan are always better
and recommended to use. For MAD data, the reflection data and f’ and
f” values can be input for all the other wavelengths collected
(1.7).
If a native data set is available, it can also be input (1.8).
The native dataset is always used for building and refinement but it
can be also used for SIRAS phasing: If only a single anomalous
dataset is input together with the native data, an option to switch
between the SIRAS and SAD phasing can be selected under the
“Important Paramaters” tab. Finally, a previously defined
cross-validation or “free” set can be input (1.9) or the
cross-validation can be turned off, otherwise the default is to
define a new set
This page (and the next one) offers options that can improve results
if default (automatically determined) values are not optimal. From
the number of protein residues per monomer, the interface will obtain
a guess for the number of molecules in the asymmetric unit and the
solvent content based on a Matthew’s coefficient analysis. These
values are only a guess and (re-)running the task with ‘correct’
values can significantly improve results.
The third pane offers advanced options, ones you will not normally
need to change for the first run. If you suspect that substructure
detection was not successful, consider running with different
resolution cutoffs (3.1) and more trials (3.2) if the results
indicate that there is an anomalous signal. For sulfur-SAD phasing,
as mentioned above, if your anomalous resolution is greater than 2.1,
adjust the minimum distance between atoms (3.3) to 1.5. Another
useful option that can improve results is more iterations in model
building.(3.4) If you wish to improve a partial model, it is
generally better to “continue” the previous job by inputting the
partially built model and phases from it by using the “CRANK2 phasing
& building” button that appears below the report from a completed
job. For all steps, the gui provides the ability to input custom
keywords for individual programs (i.e. (3.5), (3.6)). Please
look at all the documentation for individual programs (links shown
above) to see supported keywords. Multiple options can be inputted,
separated by a comma.
While the task is running, graphs are updated to indicate the
performance for each step. After and typically while each step is
being performed, a summary with statistics indicating the performance
is provided. For many graphs, a shaded gray area is shown to indicate
a region of poor statistics. For some steps, the user may click on a
button on the bottom right panel of the Results section to stop the
current step. If the button is pressed, the step will be stopped
after finishing the current iteration and the next step of the
pipeline will follow. This can be used to save time if the user is
happy with the results of the current step as indicated by the
Results report.
Figure 4: Anomalous difference to noise versus resolution¶
Besides generating the FA values, this step may also provide useful
information about the input datasets. Figure 4 shows the graph of
Dano/Sigdano - a measure of the anomalous signal strength - as a
function of resolution. The larger the number, the greater the signal
and a value of about 0.8 indicates zero signal. If unmerged data was
input, the graph of CCAnom1/2 - the anomalous difference correlation
within a dataset - versus resolution (Figure 5)
Figure 5: Anomalous difference correlation within dataset¶
is shown. Again the larger the number, the greater the signal and
values over approximately 30% indicate a significant anomalous
signal. Evaluating these graphs is useful if the substructure was not
correctly identified and the user should re-run the task by manually
inputting a resolution cutoff, or even several jobs varying the
cutoff for difficult cases with small anomalous signals. Similarly,
in the case of a MAD experiment, the signed anomalous correlation
between data sets is shown (Figure 6)
Figure 6: Anomalous difference correlation between MAD datasets¶
If there is a significant anomalous signal for different
wavelengths, a high anomalous difference correlation for them will be
present (above 30%).
A good indication that the substructure has been correctly determined
is if the threshold (i.e. CFOM in the case of SHELXD) has been
reached before the maximum number of cycles has been performed: the
report summary for the Substructure detection step will state this. A
division of trials in the Distribution of CFOM (Figure 7) is also an
encouraging indicator: the graph shows the separation of incorrect
(lower CFOM) trials from correct (higher CFOM) solutions. This
division can also be observed as clustering in a graph of CC against
CCweak where the trials (shown in circles) in the the top right
indicate correct solutions with higher values of CC and CCweak.
However, some datasets, especially those with small anomalous
signals, show little discrimination between the CFOM of solutions and
of wrong substructures. The graph of occupancies versus atom number
details the number of significant peaks that have been found and can
be used cautiously to indicate the number of substructure atoms and
protein molecules in the asymmetric unit.
A Crank2 job from SAD data will attempt to automatically improve the
substructure obtained from phasing, using anomalous likelohood
gradient maps for detection of additional anomalous scatterers. The
phasing report summary shows the number of anomalous atoms that have
been added or removed and quotes the estimated mean figure of merit
(FOM) of phases. For substructure phasing, density modification and
model building and refinement, the mean FOM is the estimated mean
cosine of the phase error: thus, a number closer to 1 is better than
a number closer to zero. Values greater than 0.3 after phasing often
indicate good phases. Yet, the average FOM is just an estimation of
the quality of the phases which may be skewed and occasionally models
can be successfully built with a low estimated FOM after phasing: for
example, this data set has a figure of merit of 0.1 after phasing,
but still leads to successful model building.
Reciprocal space information from substructure phasing can not, in
general, resolve the correct enantiomorph. However, in many cases
comparison of the quality of resulting experimental maps for both
hands can be used to distinguish the correct hand. The quality of the
experimental map can be estimated by its skewness or a related
parameter of correlation with local deviation of the map (CLD).
Furthermore, statistics from density modification of both maps, such
as FOM and contrast, can be used to determine the hand, with the
assumption that the right hand can be improved more than the wrong
one.
The Crank2 pipeline uses a combined score from comparison between the
CLD from phasing and FOM from iterations of a quick density
modification. In vast majority of cases, the right hand is correctly
identified by the scoring and a large difference between the scores
indicates that a correct hand has been chosen (thus, it may be also
used as an indirect indicator of success of the substructure
determination). However, in case of very weak experimental phases it
may happen that the map based indicators fail and the wrong hand is
picked. Such case can be often recognized by small discrimination
between the scores and the CLD values of both hands around 0.
Although such cases are rate, it is recommended to rerun the job with
the other hand
The SHELX pipeline does not attempt to determine the hand; instead,
it runs SHELXE density modification and building with both hands and
the correct hand is simply the one that leads to a model built.
However, the contrast of the map for both hands in the initial
density modification is reported in a plot (Figure X+2); a
significantly better contrast such as shown in the Figure X+2 is a
good indicator of the correct hand.
Figure 11 shows the summary from density modification, reporting if
non-crystallographic symmetry (NCS) operators were detected by Parrot
and the figure of merit (FOM) and an estimate of the bias that was
detected in the density modification (beta parameter). A value of FOM
underneath 0.35 indicates either very weak phases or a wrong
solution, ie the substructure (or rarely its hand) was not determined
correctly. A value of FOM above 0.5 indicates good phases with high
chances of model building success in the next step (except of
datasets with resolution lower than ~4A where the resolution may
become the model building bottleneck).
Figure 12: Summary and graphs of Residues, FOM and Rfactors¶
The model building summary (Figure 12) will indicate whether a
majority of the model was correctly built. In the graphs, an R-factor
under 40% and a large number of residue built in a small number of
fragments are good indication of success.
Finally, the model refinement summmary quotes the R-factors versus
the cycle number: lower R-factors (underneath 40%) are a good
indication and a small difference between the working and free
R-factors is positive. In general, the R-values should plateau - if
this hasn’t occurred, and the R-factors are still going down, more
refinement steps can further improve the model.
