[phenixbb] Restricting real-space refinement and weights to specific atom selections

[Josh Cofsky]

Thanks for that suggestion, Phil. After several weeks, I've finally found
solid answers to all my questions, which I'll write here in case anyone
reading this message is curious:

Q1: Is there any way to apply different map weights to different pieces of
the structure during refinement?
A1: Pavel tells me the answer is no—this feature is not available in Phenix
yet. The way I was able to accomplish something to a similar effect was by
performing full-model refinements at different weights. Then, in a final
refinement job, use torsion restraints from each of those output models (as
reference models) across the desired stretches of residues. Clunky, but it
solved a few problems for me.

Q2: How is the generic "weight" parameter at the bottom of the Phenix GUI
applied to the refinement?
A2: The target function is T = Tmap * weight + Trestraints. At the time of
writing this message, there is a typo in the Phenix Real-Space Refine
documentation, which incorrectly reports the target function as "T = Tmap +
weight * Trestraints" (Pavel confirmed).

Q3: Why is the PDB validation telling me I have geometric issues that were
not acknowledged by Phenix?
A3: Phil is right. The ESD for several nucleic acid bond lengths and angles
is wide as compared to the sigma tolerated by wwPDB. To fix this, I went
into the CIF files within the geostd folder and made the following changes
to the troublesome RNA and DNA nucleotide files:
*data_AD.cif, data_TD.cif, data_CD.cif, data_GD.cif:*
length:
C1'-N1, ESD=0.020
C3'-O3', ESD=0.020
angles:
O4'-C1'-N9/N1, ESD=0.30

*data_U.cif*
angles:
C5-C6-N1, ESD=0.5
C6-N1-C2, ESD=0.5

*data_C.cif*
angles:
C6-N1-C2, ESD=0.5

*data_A.cif*
angles:
N7-C8-N9, ESD=0.5


Best,
Josh


On Mon, Sep 13, 2021 at 7:23 AM Philip D. Jeffrey <pjeffrey at princeton.edu>
wrote:

> :: most of the outliers are not acknowledged as such (by Phenix), and I
> only become
> :: aware of them when the wwPDB validation tells me they are there.
>
> Check that the equilibrium values that PDB are using and the ones that
> Phenix are using are the same.  If they're not (or if the sigmas are way
> different) you're not going to be able to change the outcome just by
> changing weights.
>
> Phil Jeffrey
> Princeton
>
> ------------------------------
> *From:* phenixbb-bounces at phenix-online.org <
> phenixbb-bounces at phenix-online.org> on behalf of Josh Cofsky <
> josh.cofsky at gmail.com>
> *Sent:* Sunday, September 12, 2021 9:42 PM
> *To:* phenixbb at phenix-online.org <phenixbb at phenix-online.org>
> *Subject:* Re: [phenixbb] Restricting real-space refinement and weights
> to specific atom selections
>
> I now realize that I am confused as to what the mentioned "weight"
> parameter refers to. The Phenix RSR documentation contains a line that
> reads "T = Tmap + weight * Trestraints"—if I'm understanding correctly
> (?), this means a larger "weight" value should yield greater conformity to
> geometric restraints at the expense of model/map fit. In reality, I observe
> the opposite trend in my refinements (which is why I assumed it was a "map
> weight"). Can someone clarify what "weight" (Refinement Settings > "Other
> Options") refers to?
>
> The reason I am exploring these questions is that Phenix RSR is
> introducing bond angle outliers into a low-resolution area of a DNA
> molecule in my model—most of the outliers are not acknowledged as such (by
> Phenix), and I only become aware of them when the wwPDB validation tells me
> they are there. I assumed this was the result of overfitting to the map
> because setting "weight" to 0.0 prevented these outliers from appearing,
> but at the expense of map-model fit in the remainder of the structure
> (which is why I wanted to apply different weights to different segments of
> the model).
>
> Thanks,
> Josh
>
> On Sun, Sep 12, 2021 at 12:21 PM Josh Cofsky <josh.cofsky at gmail.com>
> wrote:
>
> Hi all,
>
> I'm using phenix.real_space_refine to refine a model into a cryo-EM map.
> My local map resolution is highly variable across the molecule, and using a
> single map weight (the parameter shown in the Refinement Settings > "Other
> Options" in the GUI) on the whole model leads to poor map-to-model fit in
> some locations and overfitting in others. Potential solutions could be to:
> 1. assign different map weights to different segments of the model (based
> on a user-specified atom selection)
> OR
> 2. perform iterative refinements with different weight settings, holding
> different (user-specified) segments of the model fixed during each
> iteration (I suspect this isn't possible based on Pavel's message here:
> http://phenix-online.org/pipermail/phenixbb/2021-May/025037.html)
>
> Is either of these approaches (or another one that accomplishes the same
> goal) possible? I am using the Phenix GUI, but if this is only possible in
> command line, and you're willing to share an example command that
> accomplishes either of these tasks, please let me know.
>
> Thanks!
> -Josh
>
> (also, apologies if this is a duplicate message...still trying to figure
> out the phenixbb listserv)
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20211019/a4b8960a/attachment.htm>