[phenixbb] help on MIR dataset

Ni, Fengyun fni at bcm.edu
Fri Oct 24 08:51:12 PDT 2014


Hi Tom,

Thank you very much for the suggestions! It takes some time for me to try all your suggestions.

Now I could get a solution for a Hg-derivative dataset at very low resolution (5 A) with the following statistics,

                               SITE  ATOM  OCCUP     X           Y            Z             B
 CURRENT VALUES:      1    Hg       0.8672  0.0160  0.7559  0.2200   48.4721
 CURRENT VALUES:      2    Hg       1.0700  0.4392  0.3669  0.2884   60.0000
 CURRENT VALUES:      3    Hg       0.0196  0.5968  0.9263  0.1392    1.0000
 CURRENT VALUES:      4    Hg       0.1662  0.1408  0.6124  0.2366    6.4178
 CURRENT VALUES:      5    Hg       0.3109  0.1818  0.3807  0.2923   56.7720

Solution # 1   BAYES-CC: 52.8 +/- 20.8 (2SD) Dataset #1   FOM: 0.62
Score type:       SKEW    CORR_RMS    NCS_OVERLAP
Raw scores:        0.23      0.85      0.00
100x EST OF CC:   52.84     49.93     31.25

The resulting map looks reasonable for the fact that several long consecutive densities could be located and it somehow looks like what my protein should be like. My native data only diffract to 3.1 A, and the model building based on the above solution (after density modification) could only build some fragment without alpha/beta secondary structures, though R/Rfree is around 0.38/0.47.

I have another Pb-derivative, it could find a following solution at 4.4 A,

                            SITE  ATOM       OCCUP     X          Y       Z         B  
 CURRENT VALUES:      1    Pb       0.2317  0.0643  0.6622  0.1180   60.0000
 CURRENT VALUES:      2    Pb       0.0534  0.9454  0.5850  0.2217    1.0000

Solution # 2  BAYES-CC: 54.3 +/- 19.2 (2SD) Dataset #1   FOM: 0.39
Score type:       SKEW    CORR_RMS    NCS_OVERLAP
Raw scores:        0.24      0.69      0.00
100x EST OF CC:   54.29     18.27     31.25

The map from Pb-derivative is not as good as that from Hg-derivative. And it could not be combined with Hg-solution when running autosol together for both datasets.

I read from phenix that there's autobuild_parallel version for building from bad maps with some options like building with BUCCANEER, does anyone have suggestions on running this? Any suggestions for what I could try next are welcome.

Thank you very much!
Fengyun


________________________________________
From: Terwilliger, Thomas Charles [terwilliger at lanl.gov]
Sent: Tuesday, August 26, 2014 10:13 AM
To: Ni, Fengyun
Cc: PHENIX user mailing list; Terwilliger, Thomas Charles
Subject: Re: [phenixbb] help on MIR dataset

Hi Fengyun,

Some things to try:

1. Are your data twinned (look at intensity moments in xtriage output)?

2. Try space groups P3121 (and P3221 and P321) in addition to P3.

3. Take your best MR solution (or any MR solutions) and calculate phases. Use those phases as input phases to autosol and it will try to find sites using those phases.  If you then get back a map that has correlation with these phases (use get_cc_mtz_mtz to check) then you probably have a solution (autosol will only use those input phases to find the sites, not in phasing).

4. Try each of your datasets separately as SAD datasets.  You can use MR SAD in autosol (similar to #2 above but using phaser to calculate phases).

All the best,
Tom T


On Aug 25, 2014, at 10:07 PM, Ni, Fengyun wrote:

> Hi everyone,
>
> I am working on a MIR dataset: one native data, three EMTS soaked data at different concentration, one PbAc2 data. My initial trial with phenix autosol did not give positive solutions. I hope I could get some information from all of you.
>
> I index and integrate the data in Mosflm and possible group is P3, though the Pointless said the most possible is P3121. But for now, I only tried P3 at different resolutions for the reason that a potential MR solution (below 20% identities, TFZ=7.1) was found in P31 space group. If P3121 is the real space group, the space is too crowed to fit my molecules unless they have some unexpected symmetry.
>
> For the difference among dataset in P3 space group, the output from SCALEIT of CCP4 gives as follows,
>
> RMS differences|RMSiso      RMSano
> ----------------------------------------
> FP-hg          |172.30      22.50
> FP-hg2         |159.20      24.07
> FP-hg3         |85.08       27.92
> FP-pb          |31.15       24.55
>
> The native data's resolution cutoff is 3.2 A, and 3.2/3.8/3.5 for different EMTS data, 3.5 for Pb data.I did notice some radiation damage for EMTS (Hg) data (large negative B-factors in SCALA), but not for Pb data. And the anomalous signal only extend to 5.6 and 4.5 A for EMTS and Pb data, respectively, as indicated by phenix.xtriage. These are all the possible heavy atom data we could obtain. So i tried all of them in phenix.autosol. The input is like follows,
>
> autosol {
>  seq_file = ../seq.dat
>  crystal_info {
>  solvent_fraction = 0.62  #determined from matthews coefficient to assure "reasonable" content in asymmetric unit.
>  }
>  native {
>    data = nat.mtz
>  }
>  deriv {
>    data = hg1.mtz
>    atom_type = Hg
>    inano = noinano *inano anoonly
>    lambda = 1.00394
>  }
>  deriv {
>    data = hg2.mtz
>    atom_type = Hg
>    inano = noinano *inano anoonly
>    lambda = 1.00394
>  }
>  deriv {
>    data = hg3.mtz
>    atom_type = Hg
>    inano = noinano *inano anoonly
>    lambda = 1.00394
>  }
>  deriv {
>    data = pb.mtz
>    atom_type = Pb
>    inano = noinano *inano anoonly
>    lambda = 0.94640
>  }
>  model_building {
>    build = False   #my structure is DNA/protein complex, so i choose to stop the building process.
> }
> }
>
> I tried at different resolutions, 3.6/4.0/4.4/4.8/5.2/5.6/6.0 A. One solution was given based on Pb data at 4.8 A with following statistics,
>
> Solution # 6  BAYES-CC: 59.7 +/- 17.2 (2SD) Dataset #4   FOM: 0.24
> Score type:       SKEW    CORR_RMS    NCS_OVERLAP
> Raw scores:        0.31      0.69      0.00
> 100x EST OF CC:   59.74     17.55     31.25
>
> Refined heavy atom sites (fractional):
>           X         Y          Z
> xyz       0.033      0.020      0.538
> xyz       0.417      0.217      0.113
> xyz       0.041      0.010      0.721
>
>
> The statistics is almost the same for its inverse, and i read from the output that the statistics for phasing other Hg datasets are very low (BAYES-CC is lower than 20) based on the above solution. The resulting map does not make much sense yet.
>
> Sorry for the long post. Any suggestion is welcome.
> Thank you very much!
> Fengyun
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/phenixbb



More information about the phenixbb mailing list