Re: [ccp4bb] Ramachandran statistics and referee responsibility

2017-03-08 Thread Pavel Afonine
Hi Evette,


(1) best practices in refining against  lower resolution data (~4 angstrom)
> to achieve the best model,
>


obtain a model that fits data best under requirement that it has zero
geometry violations (Ramachandran, Cbeta deviations, rotamers, CABLAM,
etc..).

Note, a geometry outlier (Ramachandran plot outlier, for instance) does not
necessarily mean wrong. Example of a valid outlier: page 21 here:
http://phenix-online.org/presentations/latest/pavel_validation.pdf

However, low-resolution data is unlikely to justify outliers, that's why
zero is the goal (unless there is no other strong reasons to support the
outlier).


> One might encounter a hypothetical situation where standard refinement
> approaches gave a model with poor Ramachandran statistics.  Imposing
> Ramachandran restraints gave a model with improved Ramachandran statistics
> but at the expense of higher Rfree.
>

This is likely a software issue or incorrect use by the user. Contact
refinement software developers to resolve the issue.

Pavel


Re: [ccp4bb] Ramachandran statistics and referee responsibility

2017-03-08 Thread Mark J van Raaij
I'd expect imposing Ramachandran restraints to lower the Rfree or at least the 
gap between R and Free, otherwise I would not do it. If there were genuine 
Ramachandran outliers, the restrained model might not have these included, 
while these outliers could potentially be very interesting. Of course, if there 
were many, most would probably not be genuine and genuine ones difficult to 
identify.

For your second Q, I’d vote for:

2. authors/depositors have discretion to submit/publish the model they prefer 
so long as refinement protocol is accurately described and weaknesses such as 
poor Ramachandran statistics are evident in the presentation of the data and 
not concealed

But referees should insist on them also providing the data so that whoever 
wants could re-refine the structure to their liking.

Imposing Ramachandran constraints should also be clearly described and not 
concealed…and authors and users of models refined against low-resolution data 
should be realistic about the conclusions one can draw from them.


Mark J van Raaij
Dpto de Estructura de Macromoleculas
Centro Nacional de Biotecnologia - CSIC
calle Darwin 3
E-28049 Madrid, Spain
tel. (+34) 91 585 4616
http://wwwuser.cnb.csic.es/~mjvanraaij

> On 8 Mar 2017, at 15:29, Radisky, Evette S., Ph.D.  
> wrote:
> 
> Dear all,
>  
> I have two questions.  I would like to find out the community consensus of 
> (1) best practices in refining against  lower resolution data (~4 angstrom) 
> to achieve the best model, and also (2) what manuscript referees should ask 
> for in this regard.  One might encounter a hypothetical situation where 
> standard refinement approaches gave a model with poor Ramachandran 
> statistics.  Imposing Ramachandran restraints gave a model with improved 
> Ramachandran statistics but at the expense of higher Rfree.  I would expect 
> that the model with better geometry is probably more reliable, and wonder if 
> this is the general consensus view?  I also wonder should a referee be the 
> geometry police, or should authors/depositors have discretion to 
> submit/publish the model they prefer so long as refinement protocol is 
> accurately described and weaknesses such as poor Ramachandran statistics are 
> evident in the presentation of the data and not concealed?
>  
> Thanks for your input!
>  
> Evette
>  
> Evette S. Radisky, Ph.D. 
> Associate Professor and Consultant
> Department of Cancer Biology
> Mayo Clinic Cancer Center
> _
> Griffin Cancer Research Building, Rm 310 
> 4500 San Pablo Road 
> Jacksonville, FL 32224 
> (904) 953-6372 
> http://www.mayo.edu/research/faculty/radisky-evette-s-ph-d/bio-00094471 
> 


[ccp4bb] Ramachandran statistics and referee responsibility

2017-03-08 Thread Radisky, Evette S., Ph.D.
Dear all,

I have two questions.  I would like to find out the community consensus of (1) 
best practices in refining against  lower resolution data (~4 angstrom) to 
achieve the best model, and also (2) what manuscript referees should ask for in 
this regard.  One might encounter a hypothetical situation where standard 
refinement approaches gave a model with poor Ramachandran statistics.  Imposing 
Ramachandran restraints gave a model with improved Ramachandran statistics but 
at the expense of higher Rfree.  I would expect that the model with better 
geometry is probably more reliable, and wonder if this is the general consensus 
view?  I also wonder should a referee be the geometry police, or should 
authors/depositors have discretion to submit/publish the model they prefer so 
long as refinement protocol is accurately described and weaknesses such as poor 
Ramachandran statistics are evident in the presentation of the data and not 
concealed?

Thanks for your input!

Evette

Evette S. Radisky, Ph.D.
Associate Professor and Consultant
Department of Cancer Biology
Mayo Clinic Cancer Center
_
Griffin Cancer Research Building, Rm 310
4500 San Pablo Road
Jacksonville, FL 32224
(904) 953-6372
http://www.mayo.edu/research/faculty/radisky-evette-s-ph-d/bio-00094471