Hi, Tom has answered most of your points, so I'll just follow up on a few. First, models at 44% identity are generally much more accurate than models at 31% identity, and it's possible that averaging in the poorer model is actually making the ensemble model worse. (The probability- weighted averaging done by Phaser should limit the damage, but Phaser has to make assumptions, which may not be correct, about the relative quality of the two models.) So I would do a number of things: 1. Check the superimposed models visually, to make sure that they align well (i.e. no relative domain movements). 2. If there are relative domain movements, try solving the structure with individual domains. 3. Try using the two models separately and together as an ensemble, and choose the alternative that gives the best LLG and Z-scores. 4. If none of this gives you a positive LLG, increase the RMS error estimates. You also ask about what sequence identity to provide for a model that has been modified with Chainsaw. Chainsaw (like the FFAS procedure) doesn't change the main-chain coordinates, so they will still have a level of error like you expect from the sequence identity of the template. So you should give the original sequence identity. Really sophisticated homology modeling might improve the coordinate error, so you could in that case try a run with a slightly lower RMS than the one Phaser deduces from the sequence identity for the original template. Running Phaser from CCP4 or from AutoMR should give the same result, but maybe you have slightly different versions. In some cases, a very small shift of the coordinates can change the choice of origin/ symmetry operator that places the first molecule nearest the origin, so you can get apparently different solutions. All the best, Randy On 9 May 2009, at 20:02, Tom Terwilliger wrote:
Hi Zhongli,
Here are some ideas for you:
If the LLG is negative, it may mean that the RMSD/sequence identity is not set quite right. Alternatively the number of copies could be off.
What resolution is your data? Check the R-factor of the MR model before and after a quick refinement. Are they somewhat reasonable (i.e., <48% starting, <43% after a little refinement)? Lower numbers than this would indicate that the model is pretty good.
You can run autobuild with your target sequence and MR.1.pdb and it will edit the model for you if it can...
If you have 2 different models that fit the data...check to see if they really are the same model. Easiest way:
phenix.get_cc_mtz_mtz 2FOFC_model_1.mtz 2FOFC_model_2.mtz
will see if there is a simple translation/space-group symmetry relationship that will make these 2 maps (and the models they go with) overlap.
All the best, Tom T
On May 9, 2009, at 11:00 AM, Zhongli Cui wrote:
Dear all, I am using phenix automr to solve my structure. I have 2 search models with identiy of 44% and 31% respectively aligned by ClustalW. I used thiese two pdbs as search models directly in phenix, it gave a llg of about -60 and Tfz of 7. The refined structure fit the map quite well. But I wonder if the solution is OK for the negative LLG. So what should I do next? How should I prepare the search model? Shall I edit the seach models such as remove some loops or surface side chains? If I the a search model prepared by Chainsaw was used, then what is identity I have to tell phenix? Another question, I built 2 models of the same protein and of the same x-ray diffraction data with phaser in CCP4 and Phenix. In my mind, these models should be the same. But in fact they differ from each other significantly in orientation. Is there something wrong with my calculation?
---------------- Cui Zhongli Ph.D
Dept. of Microbiology College of Life Sciences Nanjing Agricultural University 1 Weigang, Nanjing, 210095 Jiangsu Province P.R. China. Mail: [email protected] Tel: 86-25-84396753 Mob: (0)13951080529_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
------ Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills Road E-mail: [email protected] Cambridge CB2 0XY, U.K. www- structmed.cimr.cam.ac.uk