*** For details on how to be removed from this list visit the *** *** CCP4 home page http://www.ccp4.ac.uk ***
If you have access to xprep (which is a part of SHELX utility package) - it has a robust, quick, and convenient test for anomalous signal (in the subroutine set that deals with SAD, SAD, MAD, etc.). All you need is the .sca file with anomalous data in it. The proof ultimately is in the pudding. Weak (but carefully measured) anomalous signal doesn't mean bad anomalous signal :) Run an anomalous Patterson and see if anything interesting comes up... Cheers, Artem > *** For details on how to be removed from this list visit the *** > *** CCP4 home page http://www.ccp4.ac.uk *** > > > Dear Yi, > Yup- looks like no signal- > but be careful. There is something peculiar about this test as described > in the manual. Just using the "anomalous" flag does not cause Bivoet mates > to be treated separately- they are still considered equivalent in scaling > and > calculating the statistics. > > So if you had chi^2 about one scaling all the data with the anomalous > flag, > with Bivoet mates considered equivalent (but output separately), and you > still > had chi^2 about one when you rescale that dataset with Bivoet mates > considered > equivalent, that doesn't really prove there is no signal (I think). > > The manual does not say to compare chi^2 in the first and second scaling- > just that the value of chi^2 in the second round should be greater than 1. > Last paragraph: "This whole analysis assumes the error model is reasonable > and > gives chi^2 close to one when the 'Scale Anomalous' flag is used". ('scale > anomalous' unlike 'anomalous' causes bivoet mates to be treated as > non-equivalent in scaling and calculating statistics). > > The test you really want to do is to use 'scale anomalous' in the first > round, > but the author seems to fear your anomalous data is not complete enough > for > scale anomalous to work well. Give it a try, anyway! > > <If I completely misinterpreted this strategy, please someone > chime in to correct me> > > Others could better answer your final questions, but for my 2 cents: > The R values look fine, even considering redundancy is only 2 in the > second scaling. > Chi^2 is a standard statistical parameter, something like the ratio of > the obsered deviations from the mean to those expected based on your error > model. This takes into consideration the decreased redundancy with "scale > anomalous", which would improve the R-factors even in the absence of an > anomalous signal. > > If your error model were correct, and there is an anomalous signal, > you should get chi^2 > 1 with or without the anomalous flag, and > close to one with "scale anomalous" as the author indicated. > > Ed > > > > > Yi Xue wrote: >> >> Hi, all: >> I collected Cu-MAD data, I tried to detect anomalous signals using >> scalepack. >> I followed tutorial: first: scale data with 'anomalous' checked, and >> chi >> square around 1. >> second: turn off 'anomalous' flag, and merge I+ >> and >> I-. >> >> >> Output is as follows: >> > ===========================================snip============================== >> ____________________________________________________________________________ >> _____ >> >> I am confused by Chi square and R factor values. It seems that at higher >> resolution, R factor is very high, does it mean there is a big >> difference >> between I+ and I-? Is it caused by synchrotron radiation damage? >> Besides, >> what are the ideal values for R factors, should they be constant >> throughout >> all resolution shells? >> Meanwhile, Chi square values are mostly around 1, which should be >> indicative >> of no anomalous signal. Where does chi square derive from? and, what is >> the >> real meaning of it? >> >> thank you very much >> Yi >> >> > > > > >
