Re: [ccp4bb] Statistics differences

Protein crystallography Programs for crystallography X-ray detectors
Basic tutorials:
Protein Peptide

CCP4bb navigation

CCP4bb <-- 2007 <-- September 2007 <-- 07 September 2007

Previous message:
Subject: Re: Statistics differences
From: Garib Murshudov garib {- at -} YSBL {- dot -} YORK {- dot -} AC {- dot -} UK
Date: 2007-09-07

Next message:
Subject: Coot and Density Fit Analysis
From: Andrew Gulick gulick {- at -} HWI {- dot -} BUFFALO {- dot -} EDU
Date: 2007-09-07

Subject: Re: Statistics differences
From: Pavel Afonine PAfonine {- at -} LBL {- dot -} GOV
Date: 2007-09-07

Hi all,

here is very incomplete list of why the statistics is different:

- different bulk solvent models (flat mask based, Babinet, etc.);
- different parameters for mask calculation (shrink and solvent radii,
grid step) if flat bulk solvent model is used;
- different scattering factor tables (International tables, W&K,
n-gaussian);
- different structure factor calculation algorithms: FFT or direct;
- different parameters for FFT based structure factor calculations (grid
step, atom truncation radius, Badd);
- different stereo-chemistry dictionaries;
- silent (implicit or explicit) use of hydrogens;
- different minimizers;
- different X-ray refinement target functions: ML (CNS, REFMAC, PHENIX),
LS (SHELXL);
- different ways of parameterizations of ML target (sigmaa (CNS,
REFMAC), alpha&beta (PHENIX));
- different algorithms for estimation of parameters for ML;
- different algorithms for calculation of relative weights "wxc" in
Etotal = wxc*Exray+Egeom and "wxu" in Etot = wxu*Exray+Eadp;
- different ways of handling unobserved Fobs for maps calculations;
- different parameterizations for geometry restraints targets (not the
same as different dictionaries);
- different functions for ADP restraints (TNT-like, PHENIX-like, CNS-like);
- different binning schemes;
- different random seeds for SA

... oh, tired from typing, but I can easily double or triple the list above.

Yes, I agree with Garib: hopefully all these technicalities do not
affect significantly the maps.

Cheers,
Pavel.

Jacob Keller wrote:
> Dear list,
>
> I have for some time now wondered why different programs output different statistics. A low FOM
> from program A might be much better than a high FOM from program B, and so on. I wonder why, then,
> considering that statistical measures are precisely, mathematically defined, how is there any
> discrepancy? I have also wondered whether people might prefer certain programs because they are
> statistically flattering. I think in my experience I have seen even statistics like Rfree to be
> different from different programs, I think even without any refinement--so should one use that
> program last, right before composing "Table I" for publication? That seems suspicious...
>
> Jacob Keller
>

CCP4bb navigation

CCP4bb <-- 2007 <-- September 2007 <-- 07 September 2007

Previous message:
Subject: Re: Statistics differences
From: Garib Murshudov garib {- at -} YSBL {- dot -} YORK {- dot -} AC {- dot -} UK
Date: 2007-09-07

Next message:
Subject: Coot and Density Fit Analysis
From: Andrew Gulick gulick {- at -} HWI {- dot -} BUFFALO {- dot -} EDU
Date: 2007-09-07