Python-based
Hierarchical
ENvironment
for
Integrated
Xtallography
This tutorial will use some very good MIR data (Native and 5 derivatives from a rh-dehalogenase protein MIR dataset analyzed at 2.8 A) as an example of how to solve a MIR dataset with AutoSol. It is designed to be read all the way through, giving pointers for you along the way. Once you have read it all and run the example data and looked at the output files, you will be in a good position to run your own data through AutoSol.
If PHENIX is already installed and your environment is all set, then if you type:
echo $PHENIX
then you should get back something like this:
/xtal//phenix-1.3
If instead you get:
PHENIX: undefined variable
then you need to set up your PHENIX environment. See the PHENIX installation page for details of how to do this. If you are using the C-shell environment (csh) then all you will need to do is add one line to your .cshrc (or equivalent) file that looks like this:
source /xtal/phenix-1.3/phenix_env
(except that the path in this statement will be where your PHENIX is installed). Then the next time you log in $PHENIX will be defined.
To run AutoSol on the demo rh-dehalogenase data, make yourself a tutorials directory and cd into that directory:
mkdir tutorials cd tutorials
Now type the phenix command:
phenix.run_example --help
to list the available examples. Choosing rh-dehalogenase-mir for this tutorial, you can now use the phenix command:
phenix.run_example rh-dehalogenase-mir
to solve the rh-dehalogenase structure with AutoSol. This command will copy the directory $PHENIX/examples/rh-dehalogenase-mir to your current directory (tutorials) and call it tutorials/rh-dehalogenase-mir/ . Then it will run AutoSol using the command file run.sh that is present in this tutorials/rh-dehalogenase-mir/ directory.
We are going to run this MIR dataset using a parameters file "rh-dehalogenase-mir.eff". As MIR datasets have a lot of different files and heavy-atom parameters to specify, it is easiest to run MIR by editing a simple file.
This command file run.sh is simple. It says:
#!/bin/sh echo "Running AutoSol on rhodococcus dehalogenase data..." phenix.autosol rh-dehalogenase-mir.eff
The first line (#!/bin/sh) tells the system to interpret the remainder of the text in the file using the sh (or bash) -shell (sh).
The command phenix.autosol runs the command-line version of AutoSol (see Automated Structure Solution using AutoSol for all the details about AutoSol including a full list of keywords).
The second line says to run the AutoSol Wizard, and use the contents of the file rh-dehalogenase-mir.eff as parameters.
Now let’s look at the rh-dehalogenase-mir.eff parameters file. Here is the entire file:
# parameters for autosol run with rh-dehalogenase-mir Native+5 derivs
#
autosol {
seq_file = sequence.dat
crystal_info {
space_group = p21212
unit_cell = 93.796 79.849 43.108 90.000 90.000 90.00
}
native {
data = rt_rd_1.sca
}
deriv {
data = auki_rd_1.sca
atom_type = Au
sites = 5
inano = noinano *inano anoonly
lambda = 1.5418
}
deriv {
data = hgki_rd_1.sca
atom_type = Hg
sites = 5
inano = noinano *inano anoonly
lambda = 1.5418
}
deriv {
data = ndac_rd_1.sca
atom_type = Pt
sites = 5
inano = noinano *inano anoonly
lambda = 1.5418
}
deriv {
data = hgi2_rd_1.sca
atom_type = Hg
sites = 5
inano = noinano *inano anoonly
lambda = 1.5418
}
deriv {
data = smac_1.sca
atom_type = Sm
sites = 5
inano = noinano *inano anoonly
lambda = 1.5418
}
}
Notice how the brackets ({ and }) work in this file. Everything in this file after the word "autosol" that is between the opening left-bracket ({) and and the closing right-bracket (}) is part of the autosol "scope". The AutoSol wizard looks for "autosol { lots of parameters }" and interprets everything inside these brackets. Everything outside the scope "autosol" is ignored.
Within the autosol scope there are some keywords like "atom_type = Sm", these are normally one per line.
There are also additional scopes, with keywords inside them. For example the space_group and unit_cell information are inside the scope "crystal_info".
The information about the native and each derivative is in a separate scope called "native" or "deriv". You can have one native for an MIR dataset and as many derivatives as you like.
The first part of the script, with the scope "crystal_info" tells AutoSol about the cell and space-group. These values override any values read from the input data files. Next the scope "native" gives the datafile name for the native data. Then a series of "deriv" scopes give information for each of 5 derivatives. Within this "deriv" scope you can define the datafile name, the heavy-atom name, the wavelength (lambda), f_prime and f_double_prime values for that wavelength. If you specify the heavy-atom and wavelength then the AutoSol Wizard will guess the f-prime and f-double-prime values at that wavelength. However if you know these values, then you should enter them.
Note the keyword line " inano = noinano *inano anoonly " for each derivative. This is an example of how choices are specified in a parameters file. The choice with a "*" next to it is the one that is chosen (in this case, "inano" which means include anomalous differences in phasing).
The AutoSol Wizard solves MIR datasets in several step, and in the first step, the individual derivatives are all solved separately (except using difference Fouriers to phase one derivative using a solution from another). Then when all are finished all the SIR or SIRAS datasets are phased all together with SOLVE Bayesian correlated phasing. This approach works well because a substructure determination is done separately for each derivative, and if any one of them works well, then all the derivatives can be solved.
Although the phenix.run_example rh-dehalogenase-mir command has just run AutoSol from a script (run.sh), you can run AutoSol yourself from this script with the same phenix.autosol rh-dehalogenase-mir.eff command. You can also run AutoSol from a GUI. All these possibilities are described in Using the PHENIX Wizards.
Once you have started AutoSol or another Wizard, an output directory will be created in your current (working) directory. The first time you run AutoSol in this directory, this output directory will be called AutoSol_run_1_ (or AutoSol_run_1_/, where the slash at the end just indicates that this is a directory). All of the output from run 1 of AutoSol will be in this directory. If you run AutoSol again, a new subdirectory called AutoSol_run_2_ will be created.
Inside the directory AutoSol_run_1_ there will be one or more temporary directories such as TEMP0 created while the Wizard is running. The files in this temporary directory may be useful sometimes in figuring out what the Wizard is doing (or not doing!). By default these directories are emptied when the Wizard finishes (but you can keep their contents with the command clean_up=False if you want.)
Once the AutoSol wizard has started (when run from the command line), a parameters file called autosol.eff will be created in your output directory (e.g., AutoSol_run_1_/autosol.eff). This parameters file has a header that says what command you used to run AutoSol, and it contains all the starting values of all parameters for this run (including the defaults for all the parameters that you did not set).
The autosol.eff file is good for more than just looking at the values of parameters, though. If you copy this file to a new one (for example autosol_lores.eff) and edit it to change the values of some of the parameters (resolution=3.0) then you can re-run AutoSol with the new values of your parameters like this:
phenix.autosol autosol_lores.eff
This command will do everything just the same as in your first run but use only the data to 3.0 A.
While the AutoSol wizard is running, there are several places you can look to see what is going on. The most important one is the overall log file for the AutoSol run. This log file is located in:
AutoSol_run_1_/AutoSol_run_1_1.log
for run 1 of AutoSol. (The second 1 in this log file name will be incremented if you stop this run in the middle and restart it with a command like phenix.autosol run=1).
The AutoSol_run_1_1.log file is a running summary of what the AutoSol Wizard is doing. Here are a few of the key sections of the log files produced for the rh-dehalogenase MIR dataset.
Near the top of the log file you will find:
Starting AutoSol with the command:
phenix.autosol
Reading effective parameters from rh-dehalogenase-mir.eff
autosol {
atom_type = None
lambda = None
f_prime = None
f_double_prime = None
wavelength_name = peak inf high low remote
sites = None
sites_file = None
seq_file = "sequence.dat"
...
This is just a repeat of the parameters in your rh-dehalogenase-mir.eff parameters file, merged in with all the defaults for the AutoSol wizard.
The AutoSol Wizard will read in your datafiles and check their contents, printing out a summary for each one. This is done one dataset at a time (each native-derivative pair) until all have been read in. Here is the summary for the first derivative:
HKLIN ENTRY: rt_rd_1.sca FILE TYPE scalepack_no_merge_original_index GUESS FILE TYPE MERGE TYPE sca unmerged LABELS['I', 'SIGI'] CONTENTS: ['rt_rd_1.sca', 'sca', 'unmerged', 'P 21 21 2', None, None, ['I', 'SIGI']] Inverse hand of space group: P 21 21 2 HKLIN ENTRY: auki_rd_1.sca FILE TYPE scalepack_no_merge_original_index GUESS FILE TYPE MERGE TYPE sca unmerged LABELS['I', 'SIGI'] CONTENTS: ['auki_rd_1.sca', 'sca', 'unmerged', 'P 21 21 21', None, None, ['I', 'SIGI']] Inverse hand of space group: P 21 21 2 Converting the files ['rt_rd_1.sca', 'auki_rd_1.sca'] to sca format before proceeding
The input data files rt_rd_1.sca and auki_rd_1.sca are in unmerged Scalepack format. The AutoSol wizard converts everything to premerged Scalepack format before proceeding. Here is where the AutoSol Wizard identifies the format and then calls the ImportRawData Wizard:
Running import directly... WIZARD: ImportRawData
followed eventually by...
List of output files : File 1: rt_rd_1_PHX.sca File 2: auki_rd_1_PHX.sca
These output files are in premerged Scalepack format.
After completing the ImportRawData step, the AutoSol Wizard goes back to the beginning, but uses the newly-converted files rt_rd_1_PHX.sca and auki_rd_1_PHX.sca:
HKLIN ENTRY: AutoSol_run_1_/rt_rd_1_PHX.sca FILE TYPE scalepack_merge GUESS FILE TYPE MERGE TYPE sca premerged LABELS['IPLUS', 'SIGIPLUS', 'IMINU', 'SIGIMINU'] Unit cell: (93.796, 79.849, 43.108, 90, 90, 90) Space group: P 21 21 2 (No. 18) CONTENTS: ['AutoSol_run_1_/rt_rd_1_PHX.sca', 'sca', 'premerged', 'P 21 21 2', [93.796000000000006, 79.849000000000004, 43.107999999999997, 90.0, 90.0, 90.0], 2.4307589843043771, ['IPLUS', 'SIGIPLUS', 'IMINU', 'SIGIMINU']] Inverse hand of space group: P 21 21 2 Resolution from AutoSol_run_1_/rt_rd_1_PHX.sca is 2.43 HKLIN ENTRY: AutoSol_run_1_/auki_rd_1_PHX.sca FILE TYPE scalepack_merge GUESS FILE TYPE MERGE TYPE sca premerged LABELS['IPLUS', 'SIGIPLUS', 'IMINU', 'SIGIMINU'] Unit cell: (93.796, 79.849, 43.108, 90, 90, 90) Space group: P 21 21 2 (No. 18) CONTENTS: ['AutoSol_run_1_/auki_rd_1_PHX.sca', 'sca', 'premerged', 'P 21 21 2', [93.796000000000006, 79.849000000000004, 43.107999999999997, 90.0, 90.0, 90.0], 2.430806639777233, ['IPLUS', 'SIGIPLUS', 'IMINU', 'SIGIMINU']] Inverse hand of space group: P 21 21 2 Total of 2 input data files
The AutoSol Wizard uses the sequence information in your sequence file (sequence.dat) and the cell parameters and space group to guess the number of NCS copies and the solvent fraction. It will use any wavelength information you provide it to guess the values of scattering factors for the heavy-atoms. If you do not give any wavelength then a value of lambda=1.5418 (Cu K alpha) will be used.
AutoSol_guess_setup_for_scaling AutoSol Run 1 Thu Dec 18 13:34:29 2008
Setting default value of 0.5 for solvent_fraction
Setting default value of 200 for residues
Solvent fraction and resolution and ha types/scatt fact
Guessing setup for scaling dataset 1
SG P 21 21 2
cell [93.796000000000006, 79.849000000000004, 43.107999999999997, 90.0, 90.0, 90.0]
Number of residues in unique chains in seq file: 294
Unit cell: (93.796, 79.849, 43.108, 90, 90, 90)
Space group: P 21 21 2 (No. 18)
CELL VOLUME :322858.090387
N_EQUIV:4
GUESS OF NCS COPIES: 1
SOLVENT FRACTION ESTIMATE: 0.51
Total residues:294
Total Met:6
resolution estimate: 2.43
Guessing scattering factors for AU at 1.5418 A
Guesses of scattering factors for Au
Atom Lambda f' f" datafile
Native rt_rd_1_PHX.sca
Au 1.5418 -5.09 7.30 auki_rd_1_PHX.sca
The AutoSol Wizard automatically runs phenix.xtriage on each of your input datafiles to analyze them for twinning, outliers, translational symmetry, and other special conditions that you should be aware of. You can read more about xtriage in Data quality assessment with phenix.xtriage. Part of the summary output from xtriage for this dataset looks like this:
No (pseudo)merohedral twin laws were found. Patterson analyses - Largest peak height : 6.470 (corresponding p value : 0.60758) The largest off-origin peak in the Patterson function is 6.47% of the height of the origin peak. No significant pseudotranslation is detected. The results of the L-test indicate that the intensity statistics behave as expected. No twinning is suspected.
In this space group (P21 21 2) with the cell dimensions in this structure, there are no ways to create a twinned crystal, so you do not have to worry about twinning. There is also no large off-origin peak in the native Patterson, so there does not appear to be any translational pseudo-symmetry.
After all the SIR datasets are read in, the AutoSol Wizard tests for anisotropy by determining the range of effective anisotropic B values along the principal lattice directions. If this range is large and the ratio of the largest to the smallest value is also large then the data are by default corrected to make the anisotropy small (see the AutoSol web page for more discussion of the anisotropy correction). In the rh-dehalogenase case, the range of anisotropic B values is small and no correction is made:
Range of aniso B: 13.21 20.51 Not using aniso-corrected data files as the range of aniso b is only 7.3 and 'correct_aniso' is not set
Note that if any one of the datafiles in a MIR dataset has a high anisotropy, then by default all of them will be corrected for anisotropy.
The AutoSol Wizard uses SOLVE localscaling to scale MIR data. The procedure is basically to scale all the data to the native. During this process outliers that deviate from the reference values by more that ratio_out (default=10) standard deviations (using all data in the appropriate resolution shell to estimate the SD) are rejected.
The HYSS (hybrid substructure search) procedure for heavy-atom searching uses a combination of a Patterson search for 2-site solutions with direct methods recycling. The search ends when the same solution is found beginning with several different starting points. The HYSS log files are named after the datafile that they are based on and the type of differences (ano, iso) that are being used. In this rh-dehalogenase MIR dataset, the HYSS logfile for the HgKI derivative is hgki_rd_1_PHX.sca_iso_2.sca_hyss.log. The key part of this HYSS log file is:
Entering search loop: p = peaklist index in Patterson map f = peaklist index in two-site translation function cc = correlation coefficient after extrapolation scan r = number of dual-space recycling cycles cc = final correlation coefficient p=000 f=000 cc=0.186 r=015 cc=0.245 [ best cc: 0.245 ] p=000 f=001 cc=0.198 r=015 cc=0.240 [ best cc: 0.245 0.240 ] Number of matching sites of top 2 structures: 3 p=000 f=002 cc=0.174 r=015 cc=0.215 [ best cc: 0.245 0.240 ] p=001 f=000 cc=0.212 r=015 cc=0.254 [ best cc: 0.254 0.245 0.240 ] Number of matching sites of top 2 structures: 7 Number of matching sites of top 3 structures: 3 p=001 f=001 cc=0.219 r=015 cc=0.254 [ best cc: 0.254 0.254 0.245 0.240 ] Number of matching sites of top 2 structures: 8 Number of matching sites of top 3 structures: 7 Number of matching sites of top 4 structures: 3 p=001 f=002 cc=0.163 r=015 cc=0.261 [ best cc: 0.261 0.254 0.254 0.245 ] Number of matching sites of top 2 structures: 2 Number of matching sites of top 3 structures: 2 Number of matching sites of top 4 structures: 2 ... p=013 f=000 cc=0.184 r=015 cc=0.290 [ best cc: 0.299 0.291 0.290 0.290 ] Number of matching sites of top 2 structures: 6 Number of matching sites of top 3 structures: 6 Number of matching sites of top 4 structures: 6
Here a correlation coefficient of 0.5 is very good (0.1 is hopeless, 0.2 is possible, 0.3 is good) and 8 sites were found that matched in the first two tries. The program continues until 4 structures all have 6 matching sites, then ends and prints out the final correlations, after taking the top 5 sites.
Normally either hand of the heavy-atom substructure is a possible solution, and both must be tested by calculating phases and examining the electron density map and by carrying out density modification, as they will give the same statistics for all heavy-atom analysis and phasing steps. Note that in chiral space groups (those that have a handedness such as P61, both hands of the space group must be tested. The AutoSol Wizard will do this for you, inverting the hand of the heavy-atom substructure and the space group at the same time. For example, in space group P61 the hand of the substructure is inverted and then it is placed in space group P65.
The AutoSol Wizard scores heavy-atom solutions based on two criteria. The first criterion is the skew of the electron density in the map (SKEW). Good values for the skew are anything greater than 0.1. In a MIR structure determination, the heavy-atom solution with the correct hand may have a more positive skew than the one with the inverse hand. The second criterion is the correlation of local RMS density (CORR_RMS). This is a measure of how contiguous the solvent and non-solvent regions are in the map. (If the local rms is low at one point and also low at neighboring points, then the solvent region must be relatively contiguous, and not split up into small regions.) For MIR datasets, SOLVE is used for calculating phases. For a MIR dataset, a figure of merit of 0.5 is acceptable, 0.6 is fine and anything above 0.7 is very good. The scores are listed in the AutoSol log file. Here is the scoring for solution 4 (the best initial map):
AutoSol_run_1_/TEMP0/resolve.scores SKEW 0.2369411 AutoSol_run_1_/TEMP0/resolve.scores CORR_RMS 0.9131303 CC-EST (BAYES-CC) SKEW : 54.1 +/- 19.5 CC-EST (BAYES-CC) CORR_RMS : 61.5 +/- 30.6 ESTIMATED MAP CC x 100: 58.1 +/- 14.0
This is a good solution, with a high (and positive) skew (0.24), and a high correlation of local rms density (0.91)
The ESTIMATED MAP CC x 100 is an estimate of the quality of the experimental electron density map (not the density-modified one). A set of real structures was used to calibrate the range of values of each score that were obtained for phases with varying quality. The resulting probability distributions are used above to estimate the correlation between the experimental map and an ideal map for this structure. Then all the estimates are combined to yield an overall Bayesian estimate of the map quality. These are reported as CC x 100 +/- 2SD. These estimated map CC values are usually fairly close, so if the estimate is 58.1 +/- 14.0 then you can be confident that your structure is solved and that the density-modified map will be quite good.
In this case the datasets used to find heavy-atom substructures were the isomorphous differences for each derivative. For each dataset one solution was found, and that solution and its inverse were scored. The scores were (skipping extra text below):
SCORING SOLUTION 1: Solution 1 using HYSS on AutoSol_run_1_/auki_rd_1_PHX.sca_iso_1.sca. Dataset #1 SG="P 21 21 2", with 5 sites ESTIMATED MAP CC x 100: 46.8 +/- 20.9 SCORING SOLUTION 2: Solution 2 using HYSS on AutoSol_run_1_/auki_rd_1_PHX.sca_iso_1.sca and taking inverse. Dataset #1 SG="P 21 21 2", with 5 sites ESTIMATED MAP CC x 100: 32.0 +/- 32.1 SCORING SOLUTION 3: Solution 3 using HYSS on AutoSol_run_1_/hgki_rd_1_PHX.sca_iso_2.sca. Dataset #2 SG="P 21 21 2", with 5 sites ESTIMATED MAP CC x 100: 33.5 +/- 37.0 SCORING SOLUTION 4: Solution 4 using HYSS on AutoSol_run_1_/hgki_rd_1_PHX.sca_iso_2.sca and taking inverse. Dataset #2 SG="P 21 21 2", with 5 sites ESTIMATED MAP CC x 100: 58.1 +/- 14.0
In this case the best score was solution 4 (as shown above), based on the HGKI derivative and taking the inverse of the heavy-atom sites, with a ESTIMATED MAP CC x 100: 58.1 +/- 14.0 The score from the opposite hand was just 33.5+/- 37.0 and so the hand was clear.
Depending on the space group, there may be a few or infinitely many totally equivalent heavy-atom substructures for a particular native-derivative pair. These are related to each other by translations that can be thought of as offsets of the origins for the two substructures. The AutoSol Wizard identifies the allowed offsets for the space group. Then it aligns the solutions from different derivatives by finding the origin offset that maximizes the correlation of electron density in the native Fouriers for the two. Then it combines the phases from the two using addition of Hendrickson-Lattman coefficients. These combined phases are then used to score the phasing obtained by combining the two derivatives. The best combinations are iteratively combined until all available derivatives are considered and combined in an optimal fashion. Once an optimal set of derivatives and sites is found, SOLVE Bayesian correlated phasing is used to calculate a final set of native phases from the native and all the derivatives at once. Here is the best pair of derivatives from this first cycle:
Getting origin shift for 1 mapped on to 4 Keeping order of datasets for merge 2.4307589843 2.4307589843 Phases from solution 4:solve_4.mtz Phases from solution 1:solve_1.mtz Merged ha files in ha_4_1.pdb Merged files in merged_4_1.mtz FOM solution 4: 0.486 FOM solution 1: 0.415 Correlation of maps: 0.247 Ideal map correlation: 0.20169 RESULT: FOM solution 4: 0.486 FOM solution 1: 0.415 Correlation of maps: 0.247 Ideal map correlation: 0.20169 Origin offset of solution 1: [-0.5, 0.0, 0.0]
Here solutions 1 and 4 have a map correlation of 0.25, just about the same as expected (0.20) based on the FOM of the two solutions (0.49 and .44) and assuming random errors. The two solutions differ by an origin shift of 0.5 along x.
The two solutions are then phased as a group to use as the basis for density modification:
Merging a set of solutions and phasing the group with SOLVE ... PHASED SOLUTION: Solution 9 based on MIR phasing starting from solutions 4 (dataset #2) and 1 (dataset #1) ... AutoSol_run_1_/TEMP0/resolve.scores SKEW 0.1159246 AutoSol_run_1_/TEMP0/resolve.scores CORR_RMS 0.8839763 CC-EST (BAYES-CC) SKEW : 33.5 +/- 32.8 CC-EST (BAYES-CC) CORR_RMS : 57.2 +/- 34.9 ESTIMATED MAP CC x 100: 45.4 +/- 22.8
Though worse than the HGKI solution by itself, this is reasonably good solution, with a moderate a positive skew (0.12), and a good correlation of local rms density (0.88).
As the original HGKI solution was the best, it is used for density modification and finding additional sites:
SOLUTION USED TO START DEN MOD: Solution 4 using HYSS on AutoSol_run_1_/hgki_rd_1_PHX.sca_iso_2.sca and taking inverse. Dataset #2 SG="P 21 21 2" HKLIN: solve_4.mtz Testing density modification with mask_type = histograms RFACTOR: 0.2655 Best mask type so far is histograms
Heavy-atom sites are found for derivatives that are not yet solved by phasing using the current model, carrying out density modification to improve the phases, and using the improved phases along with isomorphous differences and the phase difference between the heavy atoms and the non-heavy atoms to calculate Fourier maps showing the positions of the heavy atoms. The top peaks in these maps are used as trial heavy-atom sites (if they are not already part of the heavy-atom model.
In this example solution 4 from derivative 2 is used for this phasing/density modification/Fourier procedure. Sites are are found for all the derivatives and new solutions are created and scored using the top sites for each derivative. The combinations are then tested as above, and the highest-scoring ones are kept again. The best solution found is #96:
PHASED SOLUTION: Solution 96 based on MIR phasing starting from solutions 4 (dataset #2) and 14 (dataset #1) ... AutoSol_run_1_/TEMP0/resolve.scores SKEW 0.4449184 AutoSol_run_1_/TEMP0/resolve.scores CORR_RMS 0.9306632 CC-EST (BAYES-CC) SKEW : 71.3 +/- 11.3 CC-EST (BAYES-CC) CORR_RMS : 63.3 +/- 28.2 ESTIMATED MAP CC x 100: 71.9 +/- 10.4
This is quite a good solution, with high skew (0.44) and correlation of local rms density (0.93). This solution is the best overall and is used for final phasing and density modification. Notice that it only contains two of the five derivatives. The merging procedure identifies which combinations of derivatives give the best phasing, and all the other derivatives are ignored.
Once the best heavy-atom solution or solutions are chosen based on Z-scores, these are used in a final round of phasing with SOLVE (for MIR phasing). In this case several nearly-equally-good solutions are available, and all are used in phasing, density modification and initial model-building, with the R-factor in density modification and the model-map correlation in model-building being used to identify the best solutions. The log file from phasing for solution 96 is in solve_96.prt. The heavy-atom model is refined and phases are calculated with Bayesian correlated MIR phasing. An important part of this phasing method is a statistical method of taking into account the correlation of non-isomorphism among derivatives. The extent of this correlation is listed in the solve_96.prt summary file:
SUMMARY OF CORRELATED ERRORS AMONG DERIVATIVES
DERIVATIVE: 1
CENTRIC REFLECTIONS:
DMIN: ALL 8.91 5.58 4.35 3.68 3.25 2.94 2.71 2.52
RMS errors correlated and uncorrelated with others in group:
Correlated: 54.6 67.6 58.3 57.0 65.3 58.5 34.7 38.1 37.5
Uncorrelated: 49.6 64.3 62.8 50.1 46.7 41.4 50.8 32.9 29.2
Correlation of errors with other derivs:
DERIV 2: 0.56 0.60 0.51 0.52 0.61 0.63 0.38 0.49 0.58
Here the centric reflections in derivative 1 have non-isomorphism errors related to those in derivative 2, with a correlation coefficient overall of 0.58. another way to look at this is that the RMS correlated error is 54.6 and the RMS uncorrelated (random) error is 49.6. That means that a big part of the errors are correlated, and should be treated as such. The final occupancies and coordinates are listed at the end:
SITE ATOM OCCUP X Y Z B
CURRENT VALUES: 1 Hg 0.3744 0.2772 0.2197 0.4194 6.9985
CURRENT VALUES: 2 Hg 0.4444 0.8110 0.3415 0.4388 24.1644
CURRENT VALUES: 3 Hg 0.3327 0.2629 0.2488 0.4174 21.4129
CURRENT VALUES: 4 Hg 0.0684 0.2568 0.1753 0.3437 11.1209
CURRENT VALUES: 5 Hg 0.0918 0.3076 0.2496 0.4639 39.3362
SITE ATOM OCCUP X Y Z B
CURRENT VALUES: 1 Au 0.3856 0.7926 0.3138 0.4669 19.0809
CURRENT VALUES: 2 Au 0.4300 0.2877 0.2163 0.4266 19.5977
CURRENT VALUES: 3 Au 0.3315 0.6380 0.1629 0.4836 15.1735
CURRENT VALUES: 4 Au 0.1238 0.8116 0.3356 0.4366 1.0000
CURRENT VALUES: 5 Au 0.2690 0.2873 0.2161 0.4832 7.4303
In this case the occupancies of the top sites are about 1/3, which is fine for MIR (particularly with such heavy atoms as Hg and Au).
After MIR phases are calculated with SOLVE, the AutoSol Wizard uses RESOLVE density modification to improve the quality of the electron density map. The statistical density modification in RESOLVE takes advantage of the flatness of the solvent region and the expected distribution of electron density in the region containing the macromolecule, as well as any NCS that can be found from the heavy-atom substructure. The weighted structure factors and phases (FP, PHIB) from SOLVE are used to calculate the starting map for RESOLVE, and the experimental structure factor amplitudes (FP) and MIR Hendrickson-Lattman coefficients from SOLVE are used in the density modification process. The output from RESOLVE for solution 107 can be found in resolve_96.log. Here are key sections of this output.
First, the plot of how many points in the "protein" region of the map have each possible value of electron density. The plot below is normalized so that a density of zero is the mean of the solvent region, and the standard deviation of the density in the map is 1.0. A perfect map has a lot of points with density slightly less than zero on this scale (the points between atoms) and a few points with very high density (the points near atoms), and no points with very negative density. Such a map has a very high skew (think "skewed off to the right"). This map is good, with a positive skew, though it is not perfect.
Plot of Observed (o) and model (x) electron density distributions for protein
region, where the model distribution is given by,
p_model(beta*(rho+offset)) = p_ideal(rho)
and then convoluted with a gaussian with width of sigma
where sigma, offset and beta are given below under "Error estimate."
0.03..................................................
. . .
. . .
. xxxxx .
. xo oxx .
. x . xo .
. x . xx .
p(rho) . x . xx .
. x . xxo .
. xx . xoo .
. ox . xxxo .
. ox . xx .
. ox . oxxx .
. oxx . xxx .
. xxx . oxxxxx .
0.0 xxxx........................................oxxxxx
-2 -1 0 1 2 3
normalized rho (0 = mean of solvent region)
-------------------------------------------------------------------------------
After density modification, the curve is more ideal, with a very strong positive skew:
0.03..................................................
. . .
. . .
. xxxxxx . .
. x oxx . .
. x xx .
. x xx .
p(rho) . x .oxx .
. xx . xxx .
. xx . oxxx .
. x . xxxxx .
. xx . xxxxxxx o .
. xo . oxxxxxxoo .
.xxo . xxxxxoo
xoo . xo
0.0 o................................................x
-2 -1 0 1 2 3
normalized rho (0 = mean of solvent region)
The key statistic from this RESOLVE density modification is the R-factor for comparison of observed structure factor amplitudes (FP) with those calculated from the density modification procedure (FC). In this rh-dehalogenase MIR phasing the R-factor is very low:
Overall R-factor for FC vs FP: 0.253 for 12293 reflections
An acceptable value is anything below 0.35; below 0.30 is good.
The AutoSol Wizard will create a set of free R flags indicating which reflections are not to be used in refinement. By default 5% of reflections, (up to a maximum of 2000) are reserved for this test set. If you want to supply a reflection file hires.mtz that has higher resolution than the data used to solve the structure, or has a test set already marked, then you can do this with the keyword input_refinement_file=hires.mtz. The log file tells what file is created:
Adding FreeR_flag to AutoSol_run_1_/TEMP0/solve_96.mtz Label for column with FP is 'FP' for the file AutoSol_run_1_/TEMP0/solve_96.mtz Done with adding free R set FreeR_flag added to solve_96.mtz New file: TEMP0.mtz New labin: LABIN FP=FP PHIB=PHIB FOM=FOM HLA=HLA HLB=HLB HLC=HLC HLD=HLD FreeR_flag=FreeR_flag Copying TEMP0.mtz to exptl_fobs_phases_freeR_flags_96.mtz Columns used: LABIN FP=FP PHIB=PHIB FOM=FOM HLA=HLA HLB=HLB HLC=HLC HLD=HLD FreeR_flag=FreeR_flag Checking for HL coeffs in exptl_fobs_phases_freeR_flags_96.mtz True Refinement file with freeR flags is in AutoSol_run_1_/exptl_fobs_phases_freeR_flags_96.mtz
The files to be used for model-building are listed in the AutoSol log file:
THE FILE AutoSol_run_1_/resolve_96.mtz will be used for model-building THE FILE exptl_fobs_phases_freeR_flags_96.mtz will be used for refinement
The AutoSol Wizard by default uses a very quick method to build just the secondary structure of your macromolecule, and then will try to extend that model with standard model-building. This process is controlled by the keywords helices_strands_start=True. and helices_strands_only=False . The Wizard will guess from your sequence file whether the structure is protein or RNA or DNA (but you can tell it if you want with (chain_type=PROTEIN).
If the quick model-building does not build a satisfactory model (if the correlation of map and model is less than acceptable_secondary_structure_cc=0.35), then model-building is tried again with the standard build procedure, essentially the same as one cycle of model-building with the AutoBuild Wizard (see the web page Automated Model Building and Rebuilding with AutoBuild, except that if you specify thoroughness=quick as we have in this example, the model-building is done less comprehensively to speed things up.
In this case the secondary-structure-only model-building using solution #96 produces an initial model with 179 residues built and side chains assigned to 130, and which has a model-map correlation of 0.53:
Secondary-structure model: AutoSol_run_1_/TEMP0/Build_1.pdb Log file: Build_1.log copied to Build_1.log Models to combine and extend: ['Build_1.pdb'] Using CC to score in combine_extend Model 2: Residues built=179 placed=130 Chains=8 Model-map CC=0.53 (Build_combine_extend_2.pdb) This is new best model with cc = 0.53 Refining model: Build_combine_extend_2.pdb Model: AutoSol_run_1_/TEMP0/refine_2.pdb R/Rfree=0.41/0.45
This is quite an adequate preliminary model. It is then extended in several cycles and quite a good model is produced:
Current overall_best model and map Thu Dec 18 16:21:17 2008 Working directory: /net/sunbird/scratch1/terwill/run_121808a/rh-dehalogenase-mir /AutoSol_run_1_ Model (overall_best.pdb) from: refine_8.pdb R and R-free: 0.20 0.23 Map-model CC: 0.82 Model-building logfile (overall_best.log) from: model_with_loops_9.log Model evaluation (overall_best.log_eval) from: refine_8.pdb.log_eval Map coeffs used for build (overall_best_denmod_map_coeffs.mtz) from: map_coeffs.mtz SigmaA map coeffs (overall_best_refine_map_coeffs.mtz) from: refine_map_coeffs_8.mtz
For full model-building you will want to go on and use the AutoBuild Wizard (see the web page Automated Model Building and Rebuilding with AutoBuild )
A quick summary of the results of your AutoSol run is in the AutoSol_summary.dat file in your output directory. This file lists the key files that were produced in your run of AutoSol (all these are in the output directory) and some of the key statistics for the run, including the scores for the heavy-atom substructure and the model-building and refinement statistics. These statistics are listed for all the solutions obtained, with the highest-scoring solutions first. Here is part of the summary for this rh-dehalogenase MIR dataset:
-----------CURRENT SOLUTIONS FOR RUN 1 : ------------------- *** FILES ARE IN THE DIRECTORY: AutoSol_run_1_ **** Solution # 96 BAYES-CC: 71.9 +/- 10.4 Dataset #0 FOM: 0.6 Solution 96 based on MIR phasing starting from solutions 4 (dataset #2) and 14 (dataset #1) This solution is a composite of solutions: 4 14 (Already used for Phasing at resol of 2.44) Refined Sites: 5 NCS information in: AutoSol_96.ncs_spec Experimental phases in: solve_96.mtz Experimental phases plus FreeR_flags for refinement in: exptl_fobs_phases_freeR_flags_96.mtz Density-modified phases in: resolve_96.mtz HA sites (PDB format) in: ha_96.pdb_formatted.pdb Sequence file in: sequence.dat Model in: refine_8.pdb Residues built: 283 Side-chains built: 283 Chains: 0 Overall model-map correlation: 0.82 R/R-free: 0.2/0.23 Phasing logfile in: solve_96.prt Density modification logfile in: resolve_96.log (R=0.25) Build logfile in: model_with_loops_9.log Score type: SKEW CORR_RMS Raw scores: 0.44 0.93 100x EST OF CC: 71.32 63.28 Refined heavy atom sites (fractional): deriv 1 xyz 0.277 0.220 0.419 xyz 0.811 0.342 0.439 xyz 0.263 0.249 0.417 xyz 0.257 0.175 0.344 xyz 0.308 0.250 0.464 deriv 2 xyz 0.793 0.314 0.467 xyz 0.288 0.216 0.427 xyz 0.638 0.163 0.484 xyz 0.812 0.336 0.437 xyz 0.287 0.216 0.483
Here are some of the things to look for to tell if you have obtained a correct solution:
Once you have run AutoSol and have obtained a good solution and model, the next thing to do is to run the AutoBuild Wizard. If you run it in the same directory where you ran AutoSol, the AutoBuild Wizard will pick up where the AutoSol Wizard left off and carry out iterative model-building, density modification and refinement to improve your model and map. See the web page Automated Model Building and Rebuilding with AutoBuild for details on how to run AutoBuild.
If you do not obtain a good solution, then it's not time to give up yet. There are a number of standard things to try that may improve the structure determination. Here are a few that you should always try:
For details about the AutoSol Wizard, see Automated structure solution with AutoSol. For help on running Wizards, see Using the PHENIX Wizards.