[phenixbb] Worse R factor with newer version of phenix

Tue Oct 8 20:39:42 PDT 2013

Hi Nat,

I haven’t used secondary structure restraints at all. I do have custom
restraints because I’m linking the C- and N- terminus together but that
shouldn’t affect secondary structure restraints. When I reran the refinement
with and without ADP/stereochemistry weighting it gave me similar values:

With weighting – 

Rwork 0.1573 

Rfree 0.1806

RMS (bonds) 0.0073

RMS(angles) 1.468

Clashscore: 6.1

Ramachandran: 100

Ramachandran outliers: 0

Rotamer outliers: 0

Without weighting-

Rwork: 0.1603

Rfree:  0.1840

RMS (bonds) 0.0049

RMS(angles) 1.272

Clashscore: 7.6

Ramachandran favored: 100

Ramachandran outliers: 0

Rotamer outliers: 0

That single run might have been an anomaly.

Ryan

From: phenixbb-bounces at phenix-online.org
[mailto:phenixbb-bounces at phenix-online.org] On Behalf Of Nathaniel Echols
Sent: Tuesday, October 08, 2013 5:30 PM
To: PHENIX user mailing list
Subject: Re: [phenixbb] Worse R factor with newer version of phenix

Okay, everyone experience problems like this in 1.8.4, could you please
answer the following survey:

1) Are you using secondary structure restraints?

2) If you turn off secondary structure restraints, does that fix the
problem?

3) If you leave secondary structure restraints on, but set the parameter
refinement.secondary_structure.h_bond_restraints.remove_outliers to True,
does that also yield a reasonably correct result?  (GUI users, you should be
able to just do a parameter search for "remove_outliers".)

Jan: using the files you sent me, the fix in (3) does appear to solve the
problem.

Ryan: if you have 1.3Å data and you answered yes to (1), you really
shouldn't use secondary structure restraints at high resolution.  (I'm not
sure there's any fixed cutoff at which they become irrelevant or dangerous,
but my guess would be anything below 2.0Å doesn't need them.)

Anyone else: I'm hoping you all downloaded 1.8.4 before this past weekend
(or if you're using 64-bit Linux, before this morning).  The installers
currently online have been patched to fix the problem (I think) - note
however that using parameters from restored runs will *not* work properly.

thanks,

Nat

On Tue, Oct 8, 2013 at 9:31 AM, Ryan Spencer <rspencer at uci.edu> wrote:

Hi Nat,

                I’m having a similar issue only in reverse. I’m working with
a 1.3A data set and halfway during refinement the RMS angles and bonds shoot
up (0.02/2.3) and the Rfree/Rwork drops by two points (17/15). When I don’t
use the X-ray/stereochemistry and X-ray/ADP weight optimization the
Rfree/Rwork values shoot up 10 points on the second run and the RMS
angles/bonds drop significantly. Any help would be appreciated. I’m running
the refinement on 1.8.4-1496.

Ryan Spencer

University of California, Irvine

Nowick Research Group 

From: phenixbb-bounces at phenix-online.org
[mailto:phenixbb-bounces at phenix-online.org] On Behalf Of Nathaniel Echols
Sent: Tuesday, October 08, 2013 4:24 AM
To: PHENIX user mailing list
Subject: Re: [phenixbb] Worse R factor with newer version of phenix

Whenever you see the bond and angle RMSDs blow up like this, it means that
the weight between the X-ray and geometry restraint terms was not calculated
correctly.  Probably the quickest solution is to set wxc_scale much lower -
the default is 0.5 but at this resolution I think 0.1 would be more
appropriate, and you could try setting it even lower.  (Note that if you
have weight optimization turned on this will have no effect.)  However, the
fact that you didn't see this before suggests that there may be a bug
somewhere that is specific to your data.  Could you please send us input
files and log file at bugs at phenix-online.org so one of us can take a look?
(These will be kept private, obviously.)

thanks,

Nat

On Mon, Oct 7, 2013 at 12:46 PM, Jan van Agthoven <janccp4 at gmail.com> wrote:

Hi everyone,
I've been working on a low resolution structure (3.6A) which was
previously refined with phenix  1.8_1069 (with a reference model,
strategy=individual_sites,individual_adp ). I used the same input
files and parameters and ended up with significantly different R
factors.

Start: r_work = 0.2850 r_free = 0.3189 bonds = 0.005 angles = 1.059

Final: r_work = 0.2609 r_free = 0.3160 bonds = 0.003 angles = 0.997

For the old version

and

Start: r_work = 0.2779 r_free = 0.3183 bonds = 0.006 angles = 1.141
Final: r_work = 0.2889 r_free = 0.3303 bonds = 0.021 angles = 2.737

with the new  version 1.8.4

Remarkably it also came with an increased number of C-beta deviation
(0 against 16!).

Adding tls refinement or ramachandran restrains did not help.

Now because I was using the .eff file of an old version of phenix, it
came with a long list of unused parameters. These parameters however
were all used by default.

What should I do? Should I go back to this old phenix, or is there a
way to improve the result?

Thanks,
_______________________________________________
phenixbb mailing list
phenixbb at phenix-online.org
http://phenix-online.org/mailman/listinfo/phenixbb

_______________________________________________
phenixbb mailing list
phenixbb at phenix-online.org
http://phenix-online.org/mailman/listinfo/phenixbb

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20131008/d7ec2540/attachment.htm>