Hi Raja,

here are a few comments:

- if you used TLS previously and then run a refinement without TLS (that is "strategy=..." does not include "tls"), the previous TLS refinement gets invalidated, because phenix.refine will automatically convert ANISOU into isotropic equivalent. It says it somewhere in .log file.

- the "run 1" looks better than the others indeed (judging by the R-factors and bond/angles RMSDs at this resolution).

- your original concern was the geometry problems, so using H in refinement, as well as optimizing weights, may help fixing them. So I would check the geometry too for all the runs below, and not only R-factors.

- yesterday Kendall suggested a set of good ideas to try out regarding NCS.

- I did not understand what you mean by "default extension of high resolution range in PHENIX". phenix.refine uses all reflections that it finds in your input data files, unless specified otherwise using high/low resolution cutoffs and sigma cutoffs. phenix.refine does not collect any additional data for you -:)

- at 2.9A it is very likely that you still ok to refine individual ADPs and switching to group ADP can increase R-factors. So no surprises here. I doubt that using bulk-solvent correction and anisotropic scaling increases the R-factors; probably you tried it in combination with something else, and that "something else" was the cause of increased R-values.

- are you using the latest version of PHENIX ?

Pavel.


On 11/18/09 8:47 AM, Raja Dey wrote: 
Hi Pavel,
              I finished 4 runs as you suggested. See below:
run1.
phenix.refine 1205A-p21212.mtz nov138_001.pdb ncs_groups.params main.ncs=true ncs.find_automatically=false refinement.ncs.excessive_distance_limit=None xray_data.high_resolution=2.9 refinement.main.number_of_macro_cycles=5 output.write_maps=true output.prefix=nov17
REMARK Start: r_work = 0.2803 r_free = 0.3036 bonds = 0.012 angles = 1.488
REMARK Final: r_work = 0.2438 r_free = 0.2994 bonds = 0.009 angles = 1.309
b_ave=70.09

run2. with wxc and wxu optimization
phenix.refine 1205A-p21212.mtz nov138_001.pdb ncs_groups.params main.ncs=true ncs.find_automatically=false refinement.ncs.excessive_distance_limit=None xray_data.high_resolution=2.9 refinement.main.number_of_macro_cycles=5 optimize_wxc=true optimize_wxu=true output.write_maps=true output.prefix=nov17-opt
REMARK Start: r_work = 0.2803 r_free = 0.3036 bonds = 0.012 angles = 1.488
REMARK Final: r_work = 0.2143 r_free = 0.2961 bonds = 0.032 angles = 3.281
b_ave=64.71

phenix.reduce may21_pdbset.pdb > may21_pdbset_h_added.pdb

run3.  with hydrogen
phenix.refine 1205A-p21212.mtz nov138_001_h_added.pdb ncs_groups.params main.ncs=true ncs.find_automatically=false refinement.ncs.excessive_distance_limit=None xray_data.high_resolution=2.9 refinement.main.number_of_macro_cycles=5 output.write_maps=true output.prefix=nov17-wh
REMARK Start: r_work = 0.2860 r_free = 0.3123 bonds = 0.085 angles = 1.672
REMARK Final: r_work = 0.2381 r_free = 0.3029 bonds = 0.017 angles = 1.685
b_ave=77.7

run4. with both hydrogen and optimization
phenix.refine 1205A-p21212.mtz nov138_001_h_added.pdb ncs_groups.params main.ncs=true ncs.find_automatically=false refinement.ncs.excessive_distance_limit=None xray_data.high_resolution=2.9 refinement.main.number_of_macro_cycles=5 optimize_wxc=true optimize_wxu=true output.write_maps=true output.prefix=nov17-wh-opt
REMARK Start: r_work = 0.2860 r_free = 0.3123 bonds = 0.085 angles = 1.672
REMARK Final: r_work = 0.2447 r_free = 0.3032 bonds = 0.020 angles = 1.922
b_ave=80.75

It looks output from run1 is the best, adding hydrogen and optimizing wxc and wxu did not help much. 
I have had little better Rfree and rmsd at the end of my last run.


You are right. the commands I used at the last stage looks weird. Before that run I have used "strategy=rigid_body+individual_sites+group_adp+tls" and also included bulk_solvent and scaling. The map is pretty good. But, the number of reflections used in the refinement was more than that in mtz file and that was the warning in the pdb validation summary letter. I think that's because default extension of high resolution range in PHENIX. So, I needed to run 1 macro cycle with a little lower high resolution 2.9A with the last refined pdb. I found those are the combination of commands gives the best R, Rfree and rmsd for bonds and angles. As soon as I was including bulk solven and scaling correction and/or b_group the values were increasing. Although I don't have clear explanation, but that is what I found. Any way these are very little differences and might not be so important.

I fixed the problem in dna. The problem was related with the occupancy of the 2 alternate conformations of dna. But, I still have 8 amino acids' (out of 400) phy, psy  out side Ramachandran plot. Probably this does not mater much if I I deposite this co-ordinate in PDB.
Thanks for all your suggestions though.

Regards...
Raja 


----- Original Message -----
From: Pavel Afonine <[email protected]>
Date: Tuesday, November 17, 2009 12:20 pm
Subject: Re: [phenixbb] fixing geometry for deposition
To: PHENIX user mailing list <[email protected]>

  
Hi Raja,

    
              I used the following commands at the last stage of 
      
refinement:>
    
phenix.refine 1205A-p21212.mtz nov138_001.pdb 
      
simulated_annealing=false ncs_groups.params main.ncs=true 
ncs.find_automatically=false refinement.ncs.excessive_distance_limit=None main.bulk_solvent_and_scale=false strategy=rigid_body+individual_sites xray_data.high_resolution=2.9 refinement.main.number_of_macro_cycles=1 output.write_maps=true output.prefix=nov139 --overwrite
    
  
      
The above command seems extremely weird to me for the following 
reasons:
1) Bulk-solvent correction and anisotropic scaling always has to be 
done 
(unless you are experimenting with fake data or these corrections 
have 
been already applied to Fobs, which is bad ideas anyway). By using 
"main.bulk_solvent_and_scale=false" you turn bulk-solvent 
correction and 
anisotropic scaling off.

2) By default, phenix.refine does 3 refinement macro-cycles, but 
often 
(depending how far you are from the final model) this is not enough 
for 
refinement to converge, so increasing it to 5 or so is a good idea. 
Doing just one macro-cycle 
(refinement.main.number_of_macro_cycles=1) 
does not make much sense to me. By the way, you can use shortcuts 
like 
"main.number_of_m=5".

3) You can drop this off the list since it is the default setting 
anyway: "simulated_annealing=false".

4) The refinement strategy: "strategy=rigid_body+individual_sites". 
Normally, you do the rigid body refinement at initial stages of 
refinement when your model is poor, and not at the final run. This 
is 
because the rigid-body refinement can be very rude on your model: 
bonds 
can be broken between rigid groups since no restraints is used, for 
example. Also, it's strange that the ADP refinement is turned off - 
it's 
always good to do, and at 2.8A you can still refined individual 
ADPs. 
Well, I'm not mention using TLS...

So, I would modify the above command as following:

phenix.refine 1205A-p21212.mtz nov138_001.pdb ncs_groups.params 
main.ncs=true ncs.find_automatically=false 
refinement.ncs.excessive_distance_limit=None 
xray_data.high_resolution=2.9 output.write_maps=true 
output.prefix=test 
--overwrite

and try running it as is, and with adding keywords 
"optimize_wxc=true 
optimize_wxu=true", using a model with or w/o H riding H atoms, 
etc... 
what I wrote in my previous email.

    
Regarding amino acids phy, psy..... I tried real space 
      
refinement/regularization in coot, 

This is not the same as what I suggested to try in my previous 
email. 
Anyway, I would first try the modified command above and only then 
try 
other suggested things.

Good luck,
Pavel.

_______________________________________________
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