Re: weak reflections

[Bart Hazes <bart@sherlock.mmid.ualberta.ca>]

***  For details on how to be removed from this list visit the  ***
***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***


The main issue in the ongoing discussion is that in a data set the
information content is gradually decreasing with resolution. Therefore, a
resolution cutoff is never satisfying. In the new maximum likelihood
refinement algorithms, the decrease in information content is more
approprately modelled in the sigmaa weighting.

However, from a practical perspective, more data means slower
computations. Depending on your problem and computer hardware this may or
may not be an issue. It seems useful to me to process the data to a very
optimistically high resolution but limit the resolution during the first
rounds of refinement, for instance using I/SigI as a criterium. At this
stage of refinement where the model still not very good the highest
resolution reflections would not contribute much (sigmaa weights would be
very low). Once refinement nears the final stage, you can include the
highest resolution, questionable, data. A look at the sigmaa versus
resolution plot might already indicate wheather the highest resolution
data that you just added actually contains useful information. However,
the best judgement would be to continue a few more cycles of refinement
and compare the maps based on "reasonable" data and all data. Please post
your experiences to this mailing list so that we can build up some
empirical data as to how to deal with this issue in the future.

The remark that using very weak high resolution data makes it unclear how
to interpret the reported maximum resolution of a dataset is a long
standing issue. I have plans to write a function into SFTOOLS to report an
estimate of the information content of a dataset. This function would
consider "too-weak data", "too-strong data", non-observed data, and the
sigma values. The idea would be to report the ratio of the map rms due to
noise and the map rms due to signal. When done, I will let you know though
the CCP4 mailing list.

Bart Hazes