2017 cxi merge tutorial

From cctbx_xfel
Jump to navigation Jump to search
The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.

This is an updated, worked example of data merging using cxi.merge, for presentation at the Feb 17, 2017 Berkeley Lab Serial Crystallography Workshop. Previous documentation sets are here and here.

Initial characterization

In this example, we are given integrated still-shot data collected by Danny Axford at Diamond, for P6 myoglobin, PDB code 5M3S.

  • /net/dials/raid1/aaron/zurich0038/jr_006_batches/split_reintegrated/extracted # cctbx-style integration pickles
  • /net/dials/raid1/aaron/zurich0038/jr_006_batches/sig_filter/split_reintegrated/extracted # same data, with per-image resolution cutoff during integration

Unix ls reveals 5031 *.pickle files in each directory.

Immediately there is a problem:

$ cxi.print_pickle /net/dials/raid1/aaron/zurich0038/jr_006_batches/sig_filter/split_reintegrated/extracted/*.pickle

...fails on image 0059 with a traceback; it looks like the file is corrupted.

So focus on the data without integration resolution cutoff:

$ cxi.print_pickle /net/dials/raid1/aaron/zurich0038/jr_006_batches/split_reintegrated/extracted/*.pickle

Some conclusions with the aid of grep:

  • all integration pickles have space group P6 (good)
  • distance and beam center is fixed throughout the integrated dataset
  • Unit cells are variable but do seem to cluster around 91.4 91.4 45.9 90 90 120
phenix.fetch_pdb --mtz 5m3s

Merge command file: 

#!/bin/csh -f

set effective_params = "d_min=DMIN \
data=/net/dials/raid1/aaron/zurich0038/jr_006_batches/split_reintegrated/extracted/*.pickle \
output.n_bins=10 \
pixel_size=0.172 \
backend=FS \
nproc=1 \
model=5m3s.pdb \
merge_anomalous=True \
plot_single_index_histograms=False \
scaling.algorithm=mark0 \
raw_data.sdfac_auto=False \
scaling.mtz_file=5m3s.mtz \
scaling.show_plots=False \
scaling.log_cutoff=None \
scaling.mtz_column_F=i-obs \
scaling.report_ML=True \
set_average_unit_cell=True \
rescale_with_average_cell=False \
significance_filter.apply=True \
significance_filter.min_ct=30 \
significance_filter.sigma=0.2 \
include_negatives=NEG \
postrefinement.enable=True \
postrefinement.algorithm=rs \
output.prefix=TAG"
set tag = p6m
set dmin = 2.5
set neg = True
set eff = `echo $effective_params|sed -e "s,FS,Flex,g"|sed -e "s,DMIN,$dmin,g"|sed -e "s,NEG,$neg,g"|sed -e "s,TAG,$tag,g"`

cxi.merge ${eff}
exit
cxi.xmerge ${eff}

Initial trial nproc=1 just to see if it runs. Had to fix PDB reference. Can't use *.pickle on the data= line

Scale-up trial nproc=60, no postrefinement. set the MTZ flag = jobs 

 4493 of 5031 integration files were accepted
 0 rejected due to wrong Bravais group
 11 rejected for unit cell outliers
 22 rejected for low signal
 505 rejected due to up-front poor correlation under min_corr parameter
 0 rejected for file errors or no reindex matrix

Usage: 5m3s.mtz does not contain any observations labelled [fobs, imean, i-obs]. Please set scaling.mtz_column_F to one of [iobs]. 

 File "/net/viper/raid1/sauter/proj-e/modules/cctbx_project/xfel/cxi/util.py", line 13, in is_odd_numbered
   return int(os.path.basename(file_name).split(allowable)[0][-1])%2==1

ValueError: invalid literal for int() with base 10: 'd'

Something is wrong in the ability to determine even/odd numbered-ness. Added "_extracted.pickle" in the code; had to put it first.

Table of Scaling Results: 

---------------------------------------------------------------------------------------------------------
                                      CC      N     CC     N     R     R     R   Scale  Scale    SpSig
Bin  Resolution Range  Completeness  int    int    iso   iso    int  split  iso   int    iso      Test
---------------------------------------------------------------------------------------------------------
  1 -1.0000 -  5.3861     [809/809] 80.0%     809 75.2%    805 61.0% 40.1% 52.9% 0.551 214.059 12489.8850
  2  5.3861 -  4.2749     [791/791] 54.9%     791 74.5%    791 53.0% 38.8% 49.7% 0.693 270.307 1785.4625
  3  4.2749 -  3.7345     [781/781] 65.8%     781 81.6%    781 46.5% 33.6% 40.7% 0.762 337.287 1149.4218
  4  3.7345 -  3.3930     [776/776] 63.9%     776 74.5%    776 49.3% 36.4% 48.6% 0.764 283.109  758.0388
  5  3.3930 -  3.1498     [765/765] 67.1%     765 81.9%    765 48.4% 35.6% 43.4% 0.795 338.091  533.7650
  6  3.1498 -  2.9641     [771/771] 58.6%     771 72.4%    771 49.3% 36.6% 50.7% 0.759 286.707  222.4718
  7  2.9641 -  2.8156     [765/765] 56.0%     765 72.3%    765 48.5% 35.3% 46.7% 0.765 320.954  154.5299
  8  2.8156 -  2.6930     [746/746] 63.0%     746 76.1%    746 46.4% 34.3% 42.6% 0.867 357.183  99.4430
  9  2.6930 -  2.5894     [790/790] 52.1%     790 69.4%    790 50.4% 37.4% 47.5% 0.814 314.326  113.1264
 10  2.5894 -  2.5000     [757/757] 54.9%     757 78.6%    757 52.4% 38.9% 44.4% 0.794 306.403  109.0768

All                     [7751/7751] 74.9%    7751 78.8%   7747 51.9% 36.9% 50.1% 0.680 266.538   1298.0
---------------------------------------------------------------------------------------------------------

Of course we know the data do not scale because this is a polar space group, and data must be sorted by Brehm/Diederichs method.

Breaking the indexing ambiguity

Take note of our detail instructions on Resolving an Indexing Ambiguity. Do this in three steps:

1) Generate a database of observations

step1.csh: 

#!/bin/csh -f

set effective_params = "d_min=DMIN \
data=/net/dials/raid1/aaron/zurich0038/jr_006_batches/split_reintegrated/extracted \
output.n_bins=10 \
pixel_size=0.172 \
backend=FS \
nproc=60 \
merge_anomalous=True \
plot_single_index_histograms=False \
scaling.algorithm=mark1 \
target_unit_cell=91.4,91.4,45.9,90,90,120 \
target_space_group=P6 \
raw_data.sdfac_auto=False \
include_negatives=NEG \
postrefinement.enable=False \
output.prefix=TAG"

set tag = p6m
set dmin = 2.5
set neg = False
set eff = `echo $effective_params|sed -e "s,FS,Flex,g"|sed -e "s,DMIN,$dmin,g"|sed -e "s,NEG,$neg,g"|sed -e "s,TAG,$tag,g"`

cxi.merge ${eff}

This yields 4988 of 5031 integration files accepted.

2) Sort the lattices

step2.csh: 

#!/bin/csh -f

set effective_params = "d_min=DMIN \
pixel_size=0.172 \
target_unit_cell=91.4,91.4,45.9,90,90,120 \
target_space_group=P6 \
backend=FS \
nproc=60 \
merge_anomalous=True \
output.prefix=TAG"

set tag = p6m
set dmin = 3.5
set neg = False
set eff = `echo $effective_params|sed -e "s,FS,Flex,g"|sed -e "s,DMIN,$dmin,g"|sed -e "s,NEG,$neg,g"|sed -e "s,TAG,$tag,g"`

cxi.brehm_diederichs ${eff}

BOOST crash--floating point error

^Z; kill %%

Try using d_min 3.5 instead of 2.5--still crash

Try using fewer proc; use 30 instead of 60. (increases problem size by 2**2=4) --still crash

Try nproc=15

It looks like the crash is associated with the matplotlib plot as I only experience it when I mouse-over the plot.

setenv BOOST_ADAPTBX_FPE_DEFAULT 1

14 plots total. h,k,l=2492 h,-h-k,-1=2506 total 4998

Retrieved from "http://viper.lbl.gov:8080/cctbx.xfel/index.php?title=2017_cxi_merge_tutorial&oldid=1369"