Pavel, I use same data set for refinement, so Rfree is also marked for same reflection. The only difference is that I chose “Data labels” from Phenix IMEAN or I(+) I(-). By the way, I use version 1.8.2 for refinement. Newer versions gives higher R/Rfree. Guangyu From: Pavel Afonine mailto:[email protected]> Date: Friday, February 6, 2015 at 1:46 PM To: Guangyu Zhu mailto:[email protected]>, "[email protected]mailto:[email protected]" mailto:[email protected]> Subject: Re: [phenixbb] refinement with anomalous data Hello, I have a data set for refinement at 3.8A (outshell I/sigma ~1.0). When I processed the data, I kept anomalous data I+/I-. Because at that time I think I might have anomalous signal. But later I found that the protein coordinated with wrong metal, so I don’t really have anomalous scatters. I have good model from high resolution data. When I refined the structure, I found that refined against I+/I- give lower R/Rfree (23.0%/27.0%) than against Imean (25.1%/28.3%). When I used I+/I-, I knew I doubled reflections/parameters ratio. I also let Phenix both X-ray stereochemistry/ADP weight. The refined models have similar geometry. So should I use I+/I- for refinement since it gives lower R factor? you cannot compare R-factors calculated using different sets of reflections, they are simply not comparable. So 23.0%/27.0% versus 25.1%/28.3% doesn't really mean anything in this case. If data set is not anomalous then you can use "force_anomalous_flag_to_be_equal_to=False" so that phenix.refine uses Fobs_mean = (Fobs(+) - Fobs(-))/2 in refinement. Pavel

What I meant is: I actually don’t have anomalous signal, so when I use I+/I-, kind of like I use Imean twice for each reflection. Can I do that? My guess is that the resolution is low, so double reflection/parameter ratio really benefits the refinement, although overall the weight for x-ray term is not change, because rms bond and angle are similar. Newer versions are better for my other refinement with higher resolution, but not this data set. Guangyu From: Pavel Afonine mailto:[email protected]> Date: Friday, February 6, 2015 at 2:12 PM To: Guangyu Zhu mailto:[email protected]>, "[email protected]mailto:[email protected]" mailto:[email protected]> Subject: Re: [phenixbb] refinement with anomalous data I use same data set for refinement, so Rfree is also marked for same reflection. The only difference is that I chose “Data labels” from Phenix IMEAN or I(+) I(-). What I meant is that the IMEAN set is not the same as I(+),I(-) at least because the number of reflections in IMEAN is not the same as in I(+),I(-). By the way, I use version 1.8.2 for refinement. Newer versions gives higher R/Rfree. Should be the other way around... Pavel

That’s how I think. But say if I+ is exactly same as I-, I use it twice. I got much lower R and Rfree, that means better fit. Isn’t good? And I’m also curious why. Guangyu From: Pavel Afonine mailto:[email protected]> Date: Friday, February 6, 2015 at 2:29 PM To: Guangyu Zhu mailto:[email protected]>, "[email protected]mailto:[email protected]" mailto:[email protected]> Subject: Re: [phenixbb] refinement with anomalous data If data set isn't anomalous then I don't see why you would want to refine against I+/I-. Pavel On 2/6/15 11:26 AM, Guangyu Zhu wrote: What I meant is: I actually don’t have anomalous signal, so when I use I+/I-, kind of like I use Imean twice for each reflection. Can I do that? My guess is that the resolution is low, so double reflection/parameter ratio really benefits the refinement, although overall the weight for x-ray term is not change, because rms bond and angle are similar. Newer versions are better for my other refinement with higher resolution, but not this data set. Guangyu From: Pavel Afonine mailto:[email protected]> Date: Friday, February 6, 2015 at 2:12 PM To: Guangyu Zhu mailto:[email protected]>, "[email protected]mailto:[email protected]" mailto:[email protected]> Subject: Re: [phenixbb] refinement with anomalous data I use same data set for refinement, so Rfree is also marked for same reflection. The only difference is that I chose “Data labels” from Phenix IMEAN or I(+) I(-). What I meant is that the IMEAN set is not the same as I(+),I(-) at least because the number of reflections in IMEAN is not the same as in I(+),I(-). By the way, I use version 1.8.2 for refinement. Newer versions gives higher R/Rfree. Should be the other way around... Pavel

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 What you are confusing is "reflection/parameter ratio" with "independent observation/parameter ratio". An independent observation is one where your couldn't guess the outcome of the measurement before you made it. When there is no anomalous scattering you can predict with great reliability F- once you have measured F+ so you can't count them both. This whole business of counting observations and counting parameters really doesn't work with nonlinear mathematics so it is really only a broad rule-of-thumb for us - even when you are careful in your count. Dale Tronrud On 2/6/2015 11:26 AM, Guangyu Zhu wrote:

What I meant is: I actually don’t have anomalous signal, so when I use I+/I-, kind of like I use Imean twice for each reflection. Can I do that? My guess is that the resolution is low, so double reflection/parameter ratio really benefits the refinement, although overall the weight for x-ray term is not change, because rms bond and angle are similar.

Newer versions are better for my other refinement with higher resolution, but not this data set.

Guangyu

From: Pavel Afonine mailto:[email protected]> Date: Friday, February 6, 2015 at 2:12 PM To: Guangyu Zhu mailto:[email protected]>, "[email protected] mailto:[email protected]" mailto:[email protected]> Subject: Re: [phenixbb] refinement with anomalous data

...

I use same data set for refinement, so Rfree is also marked for same reflection. The only difference is that I chose “Data labels” from Phenix IMEAN or I(+) I(-).

What I meant is that the IMEAN set is not the same as I(+),I(-) at least because the number of reflections in IMEAN is not the same as in I(+),I(-).

...

By the way, I use version 1.8.2 for refinement. Newer versions gives higher R/Rfree.

Should be the other way around...

Pavel

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb

-----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.22 (MingW32) iEYEARECAAYFAlTVGYIACgkQU5C0gGfAG13DWgCgmyDCbiXE+wzyY0biyHQCqlFG gvsAoKpa8ZGIDlBqDK0Cx+dGFmKtdix9 =AEME -----END PGP SIGNATURE-----