[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ccp4bb]: A simple question of resolution
*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
Bernard Santarsiero writes:
> I process the data to at least I/sig of 1.0, and then look at the
> completeness at the highest resolution shell. If you use some kind of
> strategy program, you should be able to get good completeness. The main
> refinement programs, like CNS, use a 1 or 2 sigma cutoff for refinement.
> So if I process the data to 1sig, then I refine to 1sig.
>
Why use a cut off in refinement? This isunecessary and a Bad Thing to do
> I also don't monkey around with the weights when processing the data.
> Trying to make the chi**2's all 1 just assumes that random errors are
> the dominating errors. I think statistical errors are at least as large
> as the random errors, and have an unknown distribution. It's probably not
> Gaussian, so you can't make the chi**2's 1. We've collected small
> molecule data on in-house detectors and at the beamline, and have some
> idea of what the true weights (and weight contributors) are.
Not monkeying around assumes the initial estimate is correct & it isn't
Jim Pflugrath writes:
> We define in our structural biology contracts that the maximum resolution
> is where the average I/sigmaI for the high resolution shell is 2 for all
> possible reflections in the shell. Please note the use of the word
> average.
>
> Furthermore, the average I/sigI is calculated before averaging
> redundancies. That is, the calculation is made on the input reflections
> and not on the averaged unique reflections that are output and used in
> subsequent calculations. The average I/sigI for the output averaged
> reflections will be a function of the input unaveraged I/sigI and the
> redundancy (i.e. multiplicity) of the input reflns.
>
This is a very fierce cutoff. If you are going to cutoff on I/sigI
_before_ averaging, how are you gaining by collecting high redundancy?
Phil