Hi Tom

When I used data in P 3(2)21, Hyss returned a CC of 0.2 for 10 sites and in P 3, I obtained a CC of 0.32 for 17 sites (out of 20 requested). From the documentation, CC 0.2 is considered possible and 0.3 is good. However, this does not translate into good models built by AutoBuild with the R factor being in the 45-50 % and low model CC (0.25). The FOM in AutoSol also seems to not reflect what it should be even if it is better than 0.4. Interestingly, when I input a dataset with low anomalous signal to 5 A (out of 3.5 A), I got an FOM of > 0.4 and Bayes-CC of 45 - but bad model.

I also tried to run Phaser/MRage in all possible SG within both point groups without any luck.

Mohamed

On Fri, Aug 7, 2015 at 10:26 PM, Terwilliger, Thomas Charles <terwilliger@lanl.gov> wrote:
Hi Mohamed,

This looks generally fine to me.  One thing that stands out though is the line:

4.0- 3.5  0.79   3.23  0.24    4001  2577

where the half-dataset CC is 0.24, much greater than for lower resolutions.  This suggests that something is systematically wrong in this resolution shell (or elsewhere).  It may be useful to cut the data at 4 A just for this reason.  As Diana points out it may be useful for other reasons as well to cut the resolution at 4 A for finding the sites.

I would suggest taking all of the datasets you have that are more or less isomorphous and putting them in a directory and then using scale_and_merge to put them all together (this is what it is designed for).  It should be able to down-weight the ones that are most different from the average.

You could also try the brute-force option in phenix.hyss to try very hard to find the sites.

All the best,
Tom T

From: mohamed noor [mohamed.noor34@gmail.com]
Sent: Friday, August 07, 2015 10:55 AM
To: Terwilliger, Thomas Charles
Cc: PHENIX user mailing list
Subject: Re: [phenixbb] Strong anomalous signal but AutoSol fails

Hi Tom

Just to check that I am on the right track, I have attached below the output from phenix.anomalous_signal using a single XDS_ASCII.HKL file from a single crystal. It is one of a few datasets, but based on Xtriage seems to be the best one.

The space group is P 3(1/2) 2 1. All datasets were collected at the peak wavelength (fluorescence scan). My initial feeling that there is a strong signal comes from CORRECT.LP. Aimless and Xtriage. IIRC, the resolution difference between the optimistic and pessimistic measurability is about 0.3-0.4 A.


Estimation of anomalous signal in a dataset

Estimating B-value for anomalous substructure as    91.2  based on
overall B-value of    75.4 (Note: you can set this with b_value_anomalous=xx)

Getting scaled data and half-datasets with scale_and_merge
Log file will be: scale.log

Files for half_dataset CC: ['TEMP4/XDS_ASCII_SG150_1800frames_reprocFriedel_0_1.sca']
Files for half_dataset CC: ['TEMP4/XDS_ASCII_SG150_1800frames_reprocFriedel_0_1.sca']
Files for half_dataset CC: ['TEMP4/XDS_ASCII_SG150_1800frames_reprocFriedel_0_1.sca']
Scaled data are in: scaled_data.mtz
Half-dataset A is in: half_dataset_a.mtz
Half-dataset B is in: half_dataset_b.mtz
Using scaled data in analysis


Setting up estimator for CC*

-------------------Summary of signal in this dataset ------------------------

       Shell
                       CCano   Nrefl Nrefl
Resolution Esqr I/sigI  half   anom   half
47.8- 7.0  0.50  19.55  0.32    2389  2360
 7.0- 6.5  0.69  10.63  0.11     621   608
 6.5- 6.0  0.87   8.47  0.08     850   818
 6.0- 5.5  0.88   7.52  0.08    1189  1141
 5.5- 5.0  0.72   5.88  0.09    1709  1588
 5.0- 4.5  0.76   4.59  0.07    2521  2247
 4.5- 4.0  0.98   3.25  0.05    3881  3268
 4.0- 3.5  0.79   3.23  0.24    4001  2577
 3.5- 3.3  1.72   1.99 -0.02    3420  2118

       Cumulative

----------------------Data quality-----------------    Best guess of expected
                                                      results of finding sites
                                                     ------ and phasing--------

                     CCano   Nrefl                    P(Substr)            
Resolution Skew Esqr  half   anom    CC* Signal  +/-     (%)       FOM*  +/-
47.8- 7.0  0.02 0.47  0.32    2389  0.51  10.1   1.3      68       0.2   0.0
47.8- 6.5  0.02 0.50  0.28    3010  0.48  10.7   1.7      72       0.2   0.1
47.8- 6.0  0.02 0.57  0.24    3860  0.43  10.8   2.8      72       0.2   0.1
47.8- 5.5  0.05 0.64  0.20    5049  0.44  12.6   2.4      78       0.2   0.1
47.8- 5.0  0.01 0.66  0.17    6758  0.39  12.8   3.3      78       0.2   0.1
47.8- 4.5  0.01 0.69  0.14    9279  0.36  13.4   3.4      82       0.2   0.1
47.8- 4.0  0.00 0.77  0.12   13160  0.33  14.3   3.6      88       0.2   0.1
47.8- 3.5  0.00 0.77  0.15   17161  0.39  18.4   3.8      98       0.3   0.1
47.8- 3.3  0.00 0.88  0.14   20581  0.31  15.7   2.1      95       0.2   0.0


On Fri, Aug 7, 2015 at 2:30 PM, Terwilliger, Thomas Charles <terwilliger@lanl.gov> wrote:
Hi Mohamed,

You might try running phenix.anomalous_signal on your data (may require finding your unmerged data and running phenix.scale_and_merge first).  This will give you an idea if you should be able to solve your SAD dataset.

See:  http://www.phenix-online.org/version_docs/1.10pre-2124/reference/anomalous_signal.html

All the best,
Tom T





From: phenixbb-bounces@phenix-online.org [phenixbb-bounces@phenix-online.org] on behalf of mohamed noor [mohamed.noor34@gmail.com]

Sent: Thursday, August 06, 2015 3:16 PM

To: PHENIX user mailing list

Subject: [phenixbb] Strong anomalous signal but AutoSol fails



















Dear developers




I have a low resolution anomalous dataset which Aimless suggests has an effective resolution to 3.3 A and anomalous signal to 3.5 A. However, SAD phasing with AutoSol is not successful with the final R factor around 50 %.





I also have another dataset collected at a remote wavelength without anomalous signal to 3 A but they are not isomorphous (> 2 A difference in c axis).






The anomalous signal comes from the ligand heme c, which is bound covalently to the protein, so its occupancy should be 1. The protein is quite small with about 120 residues. Xtriage suggests an NCS of 6 to 20 with most likely number to be 13.




Is there any reason why a reasonable solution cannot be found? There is no twinning.




I am using the latest nightly 1.10 pre2124.




Thanks.