[phenixbb] phenix.refine v.1.3 vs 1.24

[Pavel Afonine]

Hi Leo,

I can't tell what exactly happens in your case. Below are a few points...

- Version 1.24 is very old and we fixed lots of problems and made many 
improvements / features since that time. So, I would suggest to use 
version 1.3 anyway.

- I will add an option so one can turn the automatic weights adjustment 
off.

- I haven't done any systematic investigation but I've seen some cases 
where ml target produced "better" results than mlhl (unexpected) and 
inversely (as expected).

- Did you try to use riding hydrogens? In many cases this improves the 
model geometry.

- You may also want to run an automatic weight optimization procedure. 
Please note it may take a while to run (depending on model size and data 
resolution).

Cheers,
Pavel.


On 10/3/2008 9:18 AM, Leo Sazanov wrote:
> When refining the same (large) molecule in several different crystal 
> forms at about 3.1 A resolution, I noticed that in two cases when 
> experimental phase information is available, in phenix.refine-1.3 I get 
> worse Ramachandran compared to phenix.refine-1.24 - about 2% less 
> residues in most favored regions. In one case when there is no 
> experimental phase information, there is no difference between two 
> phenix versions.
> What could be the reason for that?
> These comparisons where done for cases when final RMSDs and R-factors 
> are similar (and all RMSDs calculated in v.1.3), so automatic weights 
> adjustment in v.1.3 should not have played much role. Adding to other 
> people's comments, it would be useful to be able to switch this 
> adjustment off, as sometimes one wants full control over final RMSDs, etc.
> Thanks!
>
>