116
edits
mNo edit summary |
|||
(8 intermediate revisions by 2 users not shown) | |||
Line 530: | Line 530: | ||
but there does not appear a "magic bullet" that would produce much better data than with the quick bootstrap approach. | but there does not appear a "magic bullet" that would produce much better data than with the quick bootstrap approach. | ||
== | == Trying to solve the structure == | ||
First, we repeat xscale after inserting FRIEDEL'S_LAW=FALSE into XSCALE.INP . This gives us | First, we repeat xscale after inserting FRIEDEL'S_LAW=FALSE into XSCALE.INP . This gives us | ||
Line 575: | Line 575: | ||
but it traces only about 62 residues. The density looks somewhat reasonable, though. | but it traces only about 62 residues. The density looks somewhat reasonable, though. | ||
The files [ | The files [https://{{SERVERNAME}}/pub/xds-datared/1g1c/xds-simulated-1g1c-I.mtz xds-simulated-1g1c-I.mtz] and [https://{{SERVERNAME}}/pub/xds-datared/1g1c/xds-simulated-1g1c-F.mtz xds-simulated-1g1c-F.mtz] are available. | ||
I refined against 1g1c.pdb: | I refined against 1g1c.pdb: | ||
Line 586: | Line 586: | ||
== Notes == | == Notes == | ||
=== Towards better completeness: using the first two frames === | === Towards better completeness: using the first two frames instead of only the first === | ||
We might want better (anomalous) completeness than what is given by only the very first frame of each dataset. To this end, we change in XDS.INP : | We might want better (anomalous) completeness than what is given by only the very first frame of each dataset. To this end, we change in the XDS.INP part of our script : | ||
DATA_RANGE=1 2 | DATA_RANGE=1 2 | ||
then run the script which reduces the 100 datasets. When this has finished, we insert in XSCALE.INP | |||
NBATCH=2 | NBATCH=2 | ||
after each INPUT_FILE line. The reason for this is that by default, XSCALE establishes scalefactors every | after each INPUT_FILE line (this can be easily done using <pre> awk '{print $0;print "NBATCH=2"}' XSCALE.INP > x </pre>). The reason for this is that by default, XSCALE establishes scalefactors every 5 degrees, but here we want scalefactors for every frame, because the radiation damage is so strong. This gives: | ||
NOTE: Friedel pairs are treated as different reflections. | NOTE: Friedel pairs are treated as different reflections. | ||
Line 621: | Line 621: | ||
total 165827 42032 43003 97.7% 12.8% 13.3% 162426 8.79 14.7% 15.7% 15% 0.881 16225 | total 165827 42032 43003 97.7% 12.8% 13.3% 162426 8.79 14.7% 15.7% 15% 0.881 16225 | ||
showing that the anomalous completeness, and even the quality of the anomalous signal, can indeed be increased. I doubt, however, that going to three or more frames would improve things even more. | showing that the anomalous completeness, and even the quality of the anomalous signal, can indeed be increased. I doubt, however, that going to three or more frames would improve things even more. | ||
The MTZ files are at [https://{{SERVERNAME}}/pub/xds-datared/1g1c/xds-simulated-1g1c-F-2frames.mtz] and [https://{{SERVERNAME}}/pub/xds-datared/1g1c/xds-simulated-1g1c-I-2frames.mtz], respectively. They were of course obtained with XDSCONV.INP: | |||
INPUT_FILE=temp.ahkl | |||
OUTPUT_FILE=temp.hkl CCP4_I | |||
for the intensities, and | |||
INPUT_FILE=temp.ahkl | |||
OUTPUT_FILE=temp.hkl CCP4 | |||
for the amplitudes. In both cases, after xdsconv we have to run | |||
<pre> | |||
f2mtz HKLOUT temp.mtz<F2MTZ.INP | |||
cad HKLIN1 temp.mtz HKLOUT output_file_name.mtz<<EOF | |||
LABIN FILE 1 ALL | |||
END | |||
EOF | |||
</pre> | |||
Using the default (see above) phenix.refine job, I obtain against the [https://{{SERVERNAME}}/pub/xds-datared/1g1c/xds-simulated-1g1c-F-2frames.mtz MTZ file with amplitudes]: | |||
Start R-work = 0.3434, R-free = 0.3540 | |||
Final R-work = 0.2209, R-free = 0.2479 | |||
and against the [https://{{SERVERNAME}}/pub/xds-datared/1g1c/xds-simulated-1g1c-I-2frames.mtz MTZ file with intensities] | |||
Start R-work = 0.3492, R-free = 0.3606 | |||
Final R-work = 0.2244, R-free = 0.2504 | |||
so: '''better R-free is obtained from better data.''' | |||
The statistics from SHELXD and SHELXE don't look better - they were already quite good with a single frame per dataset. The statistics printed by SHELXE (for the correct hand) are: | |||
... | |||
<wt> = 0.300, Contrast = 0.591, Connect. = 0.740 for dens.mod. cycle 50 | |||
Estimated mean FOM and mapCC as a function of resolution | |||
d inf - 3.98 - 3.13 - 2.72 - 2.47 - 2.29 - 2.15 - 2.04 - 1.95 - 1.87 - 1.81 | |||
<FOM> 0.601 0.606 0.590 0.570 0.538 0.542 0.521 0.509 0.529 0.498 | |||
<mapCC> 0.841 0.813 0.811 0.786 0.763 0.744 0.727 0.740 0.761 0.722 | |||
N 2289 2303 2334 2245 2289 2330 2299 2297 2429 2046 | |||
Estimated mean FOM = 0.551 Pseudo-free CC = 59.42 % | |||
... | |||
Site x y z h(sig) near old near new | |||
1 0.7375 0.6996 0.1537 20.4 1/0.06 2/15.05 6/21.38 3/21.54 5/22.03 | |||
2 0.7676 0.7231 0.3419 18.8 3/0.13 5/12.15 1/15.05 3/21.34 4/22.43 | |||
3 0.5967 0.4904 -0.0067 17.2 4/0.10 4/4.90 6/4.94 2/21.34 1/21.54 | |||
4 0.5269 0.5194 -0.0498 17.1 2/0.05 3/4.90 6/7.85 2/22.43 1/22.96 | |||
5 0.4857 0.6896 0.4039 -4.8 3/12.04 2/12.15 1/22.03 3/22.55 2/22.85 | |||
6 0.5158 0.4788 0.0406 4.7 5/1.45 3/4.94 4/7.85 1/21.38 5/23.30 | |||
=== Why this is difficult to solve with SAD phasing === | === Why this is difficult to solve with SAD phasing === | ||
Line 629: | Line 675: | ||
"This crystal form contains two molecules in the asymmetric unit. They are related by a noncrystallographic two-fold axis, parallel to the crystallographic b axis, located at X = 0.25 and Z = 0.23. This arrangement results in a peak in the native Patterson map at U = 0.5, V = 0, W = 0.47 of peak height 26 σ (42% of the origin peak)." | "This crystal form contains two molecules in the asymmetric unit. They are related by a noncrystallographic two-fold axis, parallel to the crystallographic b axis, located at X = 0.25 and Z = 0.23. This arrangement results in a peak in the native Patterson map at U = 0.5, V = 0, W = 0.47 of peak height 26 σ (42% of the origin peak)." | ||
Unfortunately | Unfortunately, the arrangement of substructure sites has (pseudo-)translational symmetry, and may be related to a centrosymmetric arrangement. Indeed, the original structure was solved using molecular replacement. | ||
Using the four sites as given by SHELXE (and default parameters otherwise), I obtained from the [http://cci.lbl.gov/cctbx/phase_o_phrenia.html cctbx - Phase-O-Phrenia server] the following | |||
Plot of relative peak heights: | |||
|* | |||
|* | |||
|* | |||
|* | |||
|** | |||
|** | |||
|*** | |||
|**** | |||
|****** | |||
|************ | |||
|******************** | |||
|***************************** | |||
|********************************* | |||
|*************************************** | |||
|************************************************ | |||
|************************************************************ | |||
|************************************************************ | |||
|************************************************************ | |||
|************************************************************ | |||
|************************************************************ | |||
------------------------------------------------------------- | |||
Peak list: | |||
Relative | |||
height Fractional coordinates | |||
97.8 0.01982 0.49860 -0.00250 | |||
80.2 0.17362 0.71758 0.83714 | |||
71.5 0.02405 0.53538 0.48365 | |||
63.9 -0.00511 0.07044 0.50289 | |||
62.1 0.02410 0.94827 0.48807 | |||
61.3 0.16922 0.28605 0.15985 | |||
56.3 0.12047 0.50910 0.43665 | |||
55.9 0.21871 0.26331 0.30008 | |||
55.7 0.10931 0.47245 0.53659 | |||
53.0 0.22211 0.23746 0.39503 | |||
52.9 0.03449 -0.00661 0.98264 <------ this peak is close to the origin | |||
52.5 0.06905 0.02372 0.05632 <------ this one, too | |||
... | |||
so the strongest peak corresponds to the translation of molecules (0,0.5,0) but the origin peak is at 1/2 of that size, which appears significant. | |||
=== Finally solving the structure === | |||
After thinking about the most likely way that James Holton used to produce the simulated data, I hypothesized that within each frame, the radiation damage is most likely constant, and that there is a jump in radiation damage from frame 1 to 2. Unfortunately for this scenario, the scaling algorithm in CORRECT and XSCALE was changed for the version of Dec-2010, such that it produces best results when the changes are smooth. Therefore, I tried the penultimate version (May-2010) of XSCALE - and indeed that gives significantly better results: | |||
NOTE: Friedel pairs are treated as different reflections. | |||
SUBSET OF INTENSITY DATA WITH SIGNAL/NOISE >= -3.0 AS FUNCTION OF RESOLUTION | |||
RESOLUTION NUMBER OF REFLECTIONS COMPLETENESS R-FACTOR R-FACTOR COMPARED I/SIGMA R-meas Rmrgd-F Anomal SigAno Nano | |||
LIMIT OBSERVED UNIQUE POSSIBLE OF DATA observed expected Corr | |||
8.05 1922 467 476 98.1% 4.0% 5.8% 1888 22.37 4.5% 2.5% 84% 1.952 142 | |||
5.69 3494 864 882 98.0% 4.7% 6.0% 3429 20.85 5.4% 3.2% 77% 1.707 297 | |||
4.65 4480 1111 1136 97.8% 5.1% 5.9% 4395 21.13 5.8% 3.3% 68% 1.518 406 | |||
4.03 5197 1325 1357 97.6% 5.3% 6.0% 5101 20.57 6.1% 3.8% 48% 1.280 499 | |||
3.60 5915 1500 1533 97.8% 6.0% 6.3% 5803 19.99 6.9% 4.1% 41% 1.169 572 | |||
3.29 6601 1657 1694 97.8% 6.5% 6.5% 6476 19.42 7.5% 4.6% 27% 1.066 634 | |||
3.04 7080 1789 1830 97.8% 7.6% 7.2% 6948 17.50 8.7% 5.4% 23% 1.037 693 | |||
2.85 7682 1945 1979 98.3% 8.8% 9.0% 7528 14.75 10.1% 7.0% 15% 0.935 750 | |||
2.68 8099 2062 2100 98.2% 11.0% 11.1% 7933 12.81 12.7% 9.1% 13% 0.881 795 | |||
2.55 8351 2155 2201 97.9% 13.3% 13.7% 8178 11.16 15.4% 11.0% 12% 0.872 836 | |||
2.43 9195 2327 2376 97.9% 16.5% 17.2% 9003 9.49 19.0% 15.1% 8% 0.838 904 | |||
2.32 9495 2377 2428 97.9% 19.8% 20.3% 9304 8.62 22.7% 17.3% 4% 0.818 934 | |||
2.23 9936 2498 2551 97.9% 20.8% 21.7% 9751 8.30 23.9% 17.5% 4% 0.830 987 | |||
2.15 10217 2577 2622 98.3% 23.3% 24.0% 9990 7.74 26.7% 19.2% 4% 0.814 998 | |||
2.08 10710 2704 2766 97.8% 27.1% 28.6% 10506 6.82 31.1% 23.5% 5% 0.812 1071 | |||
2.01 10899 2777 2839 97.8% 28.1% 29.2% 10648 6.46 32.3% 25.0% 6% 0.813 1059 | |||
1.95 11361 2878 2937 98.0% 34.4% 35.5% 11134 5.55 39.5% 30.3% 3% 0.780 1136 | |||
1.90 11639 2941 3000 98.0% 40.5% 41.5% 11403 4.88 46.6% 35.9% 0% 0.787 1163 | |||
1.85 12020 3068 3123 98.2% 52.2% 55.1% 11752 3.79 60.0% 47.4% 6% 0.775 1195 | |||
1.80 11506 3003 3173 94.6% 60.8% 64.8% 11229 3.23 70.1% 58.8% 0% 0.765 1148 | |||
total 165799 42025 43003 97.7% 11.7% 12.3% 162399 10.07 13.5% 14.8% 17% 0.908 16219 | |||
Using these data (stored in [https://{{SERVERNAME}}/pub/xds-datared/1g1c/xscale.oldversion]), I was finally able to solve the structure (see screenshot below) - SHELXE traced 160 out of 198 residues. All files produced by SHELXE are in [https://{{SERVERNAME}}/pub/xds-datared/1g1c/shelx]. | |||
[[File:1g1c-shelxe.png]] | |||
It is worth mentioning that James Holton confirmed that my hypothesis is true; he also says that this approach is a good approximation for a multi-pass data collection. | |||
However, generally (i.e. for real data) the smooth scaling (which also applies to absorption correction and detector modulation) gives better results than the previous method of assigning the same scale factor to all reflections of a frame; in particular, it correctly treats those reflections near the border of two frames. | |||
Phenix.refine against these data gives: | |||
Start R-work = 0.3449, R-free = 0.3560 | |||
Final R-work = 0.2194, R-free = 0.2469 | |||
which is only 0.15%/0.10% better in R-work/R-free than the previous best result (see above). | |||
This example shows that it is important to | |||
* have the best data available if a structure is difficult to solve | |||
* know the options (programs, algorithms) | |||
* know as much as possible about the experiment |