[phenixbb] strategy suggestion ???

Pavel Afonine pafonine at lbl.gov
Tue May 31 07:26:57 PDT 2011


Hi,

> 4,phenix.refine with the following command line:
>
> phenix.refine protein.1.mtz protein.pdb \
> strategy=individual_sites+individual_sites_real_space+individual_adp+tls \
> simulated_annealing=true simulated_annealing.start_temperature=1000 \
> simulated_annealing.cool_rate=10 main.number_of_macro_cycles=5 \
> ordered_solvent=true refinement.input.xray_data.labels="F,SIGF" \
> output.prefix=myprotein

- as Fred mentioned, SA is typically done at the beginning of structure 
refinement when the structure is supposed to have gross errors. At final 
stages of refinement SA may be rather harmful than helpful.

- you did not define TLS groups which means the whole molecule is 
treated as one TLS group. Most of the time it is suboptimal. Use 
phenix.find_tls_groups to determine TLS groups automatically. You need 
to re-run this command every so often to update TLS selections as your 
model improves.

- what is the resolution? Typically, refinement strategy is a function 
of both: data quality (resolution & completeness) and current model 
quality. This may significantly change refinement strategy.

- customization for SA seems a bit odd to me (cool_rate=10 & 
start_temperature=1000).

- water picking should probably be not run at early stages of refinement 
if you do not use fix_rotamers option (if you use fix_rotamers then it 
is safer to use ordered_solvent=true).

> 6,Chage the strategy line to
> strategy=individual_adp adp.individual.isotropic=all \
> to remove the ANISOU lines in pdb file.

This seems TOTALLY WRONG. To understand why, see:
http://www.phenix-online.org/presentations/latest/pavel_validation.pdf
http://www.phenix-online.org/presentations/latest/pavel_refinement_general.pdf

> Any suggestion for this workflow ? or how do you always deal with the 
> similar case ? cause I'm new about this field,maybe I did something 
> stupid.

See above.

> Ps, the value of Rwork and Rfree are too close, like,
> Final: r_work = 0.1741 r_free = 0.1817 bonds = 0.036 angles = 1.818
> Is that reasonable ?

Pages #36-42 here:
http://www.phenix-online.org/presentations/latest/pavel_validation.pdf
contain the answer to this question.

Also, you may want to run AutoBuild (phenix.autobuild) to automatically 
rebuild your model which should reduce the amount of manual model building.

Cheers,
Pavel.



More information about the phenixbb mailing list