Dear John, many thanks  for the reference. Discussions of spatial 
autocorrelation measures seem to come up mostly in the econometrics and 
geography literature. An expert (I am not ) will no doubt supply better 
citations, but the chapters by Haining and Getis in "Handbook of Applied 
Spatial Analysis"
http://www.springer.com/economics/regional+science/book/978-3-642-03646-0
may be of use. There is  a discussion of testing for white noise with no 
spatial dependence in the context of image analysis (courtesy of Peyman 
Milanfar ) in section III of
http://www.lx.it.pt/~mtf/Almeida_Figueiredo_TIP2013.pdf

What I was asking about is  a fairly simple-minded application of these 
tests. Density from calculated structure factors for a  10A cube P1 unit 
cell with either O or Na at the center shows clear difference up to 2A 
from the peak:
http://diablo.ucsc.edu/~afyfe/navso/navso.pdf
Pearson correlation ignores position and only considers density values. 
For distinguishing Na and water amidst the noise of  a real map 
however,  it might (or not..) be helpful to test whether a cluster of 
positive/negative values near the peak differs from what can be expected 
randomly in the map.
Best,
Alastair Fyfe

  On 11/09/2013 11:27 PM, Jrh wrote:
> Dear Alastair,
> This reference:-
> http://dx.doi.org/10.1107/S0907444903004219
> seems related to your email input below.
> I would be grateful though to be guided with a couple of references from you on the topics you raise.
> Thankyou in anticipation,
> Best wishes,
> John
>
> Prof John R Helliwell DSc FInstP CPhys FRSC CChem F Soc Biol.
> Chair School of Chemistry, University of Manchester, Athena Swan Team.
> http://www.chemistry.manchester.ac.uk/aboutus/athena/index.html
>   
>   
>
> On 9 Nov 2013, at 20:30, Alastair Fyfe <[log in to unmask]> wrote:
>
>> A related question on this topic: calculated density curves  drop off at different rates even for isoelectronic ions/water. Thus the neighborhood of a mismodeled peak in the error map would be expected to show detectable, non-random, spatial dependence. On the other hand the neighborhood of a well-modeled peak  should be indistinguishable from white noise. Though there are statistics (Moran's I, Geary's C ) for testing spatial effects  in variable correlation that could be applied to  DFc/ 2mFo-DFc correlation,  I haven't seen them applied to this problem. Can anyone suggest a relevant reference? This seems a useful adjunct to bond-valence/non-bonding contact methods.
>> thanks,
>> Alastair  Fyfe
>>
>> On 11/09/2013 03:24 AM, Robbie Joosten wrote:
>>> Hi Ed,
>>>
>>> WHAT_CHECK checks water that may be ions and also checks the identity of ions already built. The check my metal server is also very good for final validation of ions.
>>>
>>> Cheers,
>>> Robbie
>>>
>>> Sent from my Windows Phone
>>> ________________________________
>>> Van: Edward A. Berry
>>> Verzonden: 9-11-2013 7:29
>>> Aan: [log in to unmask]
>>> Onderwerp: Re: [ccp4bb] monovalent cation binding sites
>>>
>>> Thanks, all!
>>> Ed
>>>
>>>
>>> Nat Echols wrote:
>>>> In the latest Phenix:
>>>>
>>>> mmtbx.water_screen model.pdb data.mtz elements=NA,K
>>>>
>>>> The data are required right now, but I could be convinced to make that optional.
>>>>
>>>> -Nat
>>> Diana Tomchick wrote:
>>>> There's a command in coot that identifies "waters" that have an unusually high coordination number. You then need to manually inspect the electron density map and bond lengths, atom type, etc.
>>>>
>>>> Diana
>>> Shekhar Mande wrote:
>>>> Ed, I dont about monovalent metals (they are typically liganded by hydroxyls of Ser/Thr,
>>>> or the main chain carbonyls).  But we did an analysis of divalent metals in proteins, and
>>>> found several instances, where crystallographers might have mistaken a metal to be a
>>>> water.  Thus, in the PDB, what is reported to be a water, might actually turn out to be a
>>>> metal!  Some of our sites have used to predict functions of proteins, where enzyme assays
>>>> required addition of metals, and hence, I am gratified that it is useful!  I am enclosing
>>>> a PDF with this.
>>>>
>>>> We also have a server.
>>>>
>>>> Shekhar
>>> Dunten, Pete W. wrote:
>>>> Ed,   O had a command
>>>> that scrutinized waters
>>>> and helped find metals
>>>> modeled as water.
>>>>
>>>> Victor Lamzin has a program
>>>> whose name I'be momentarily
>>>> forgotten which gives plots of
>>>> e-density at atomic centers
>>>> versus B-factor,  for each atom
>>>> type.  Points off the lines are
>>>> candidates for incorrectly
>>>> modeled metals.
>>>>
>>>> Pete
>>> Dunten, Pete W. wrote:
>>>> See attached and the reference noted therein.
>>>>
>>>> Best wishes, Pete
>>>> ________________________________________
>>> Parthasarathy Sampathkumar wrote:
>>>> Hi Ed,
>>>>
>>>> WASP analyse water molecules in high-resolution protein structure to check if some of
>>>> those could be metal ions. WASP could be run as a part of STAN server.
>>>> STAN - the STructure ANalysis server from USF (
>>>> http://xray.bmc.uu.se/cgi-bin/gerard/rama_server.pl )
>>>>
>>>> One could also identity potential metal ions within COOT as well.
>>>>
>>>> HTH,
>>>> Best Wishes,
>>>> Partha
>>>>
>>>>
>>>>
>>>>
>>>> On Fri, Nov 8, 2013 at 9:09 PM, Edward A. Berry <[log in to unmask]
>>>> <mailto:[log in to unmask]>> wrote:
>>>>
>>>>      Is there a server or program to predict binding sites for monovalent metal ions?
>>>>      Ideally should work with just the protein structure, but a program that sorts through
>>>>      the waters in a high resolution structure and tells which are likely to be K+ or Na+
>>>>      would also be of interest.
>>>>
>>>>      Ed
>>>>
>>>>