Hi Bernhard,

The rmsZ only says something about the deviations from 'ideal' values given 
their esd. It is by no means a strong validation criterion. Values over 1.000 
only say that your distribution is wider than expected (if you have no serious 
individual outliers). So by changing target esds to get a lower number you are 
only kidding yourself. For most bonds and angles we have the E&H or the 
Parkinson targets and esds not using those should be in the methods section of 
the paper. Note that the esds are not the same for different bonds/angles. 
Listing all the target values and esds is rather impractical, but it is indeed 
the raw data you need to reproduce the number.
My problem with rmsds the way they are used in Table 1 is that they are based 
on values with different esds so they cannot be compared in different 
structures. An even bigger problem is that some people consider them strong 
validation criteria and go for specific values that sometimes are even 
resolution dependent. That is a reason to avoid mentioning rmsds in this 
context.
So I agree that rmsZ values are not the final answer, but they sure beat rmsds 
when it comes to making a Table 1.

Cheers,
Robbie

Sent with my Windows Phone
________________________________
Van: [email protected]<mailto:[email protected]>
Verzonden: ‎6-‎2-‎2015 17:19
Aan: 'Robbie Joosten'<mailto:[email protected]>; 
[email protected]<mailto:[email protected]>
Onderwerp: RE: [ccp4bb] Bin R and Rfree values

>please report the bond length and angle rmsZ next to (or instead of) the
rmsd in Table 1.



'Instead' alone is also not enough - if in a  normalized statistic you do
not know what you normalized against - that is, the variance or esd of the
target distribution - you do not know what Z means. For example, I could
override the default set of my target rmsd in Refmac to what I want in order
to manipulate the resulting model rmsZ.

The absolute rmsd value alone also does not tell you about the underlying
target variance. A bond length rmsd of 0.04A would make you cringe but only
because you assume you know the unrestrained or true target rmsd of about
0.02A). One reason why the restraint target table in the PDB header is
useful.



I think both values, rmsd + rmsZ, or rmsd and target variance or esd, are
necessary  for getting the complete picture.



Best, BR





Reply via email to