 |
 |
>Not all counts are created equal. Your metric would tend to favour datasets with high background scatter. Yes, there *is* more information present, but not all of it is useful. I meant the number of counts in the spots. JPK -- David On 5 November 2015 at 02:11, Keller, Jacob <[log in to unmask]<mailto:[log in to unmask]>> wrote: I thought of another corollary to this way of thinking: one could quantify precisely the information content of the dataset by simply reporting the number of counts. And one could normalize this to what one would expect for a given size crystal. Actually, why don't we do this already? JPK -----Original Message----- From: CCP4 bulletin board [mailto:[log in to unmask]<mailto:[log in to unmask]>] On Behalf Of Keller, Jacob Sent: Wednesday, November 04, 2015 6:06 PM To: [log in to unmask]<mailto:[log in to unmask]> Subject: Re: [ccp4bb] New Rule for PADs When all data are zeros and ones, the answer would seem to be to start considering intensities just as probabilities. To derive errors, one could look at how the probabilities for each pixel/voxel/roxel vary as a function of the window of integration. To address the indexing issue, one could do what James Holton suggested many posts ago in a different thread, which is simply to average equivalent frames from successive sweeps, essentially recapitulating the more usual type of datasets. Why, then, do the new low-dose-rate regimen? Because after indexing the data with the averaged images, one can revert to the raw images, measuring intensities (or probabilities) within the spot predictions. Doing this procedure should yield much better estimates of errors, both within spots and in the background, which might be beneficial. One caveat, however, would be the necessity of accurate goniometers, so that one would be sure that one is averaging the correct frames together. I would further note that nothing about the collection process has changed except for the dose-rate at which it is sampled, so if one wants to return to the more familiar images, one can do that quite easily by averaging or summing the constituent images. But what I find compelling and interesting about this new type of approach is that there are three potential paradigm-shifts: 1. One could now consider the "unit" of datasets to be the 3D reciprocal space volume which changes as a function of dose. 2. Intensities could be considered probabilities 3. Datasets could be treated as "processes" or "evolutions" as a function of dose, and maybe eventually our models will reflect that. All the best, Jacob -----Original Message----- From: CCP4 bulletin board [mailto:[log in to unmask]<mailto:[log in to unmask]>] On Behalf Of Francis Reyes Sent: Wednesday, November 04, 2015 4:51 PM To: [log in to unmask]<mailto:[log in to unmask]> Subject: Re: [ccp4bb] New Rule for PADs On Wed, Nov 4, 2015 at 7:45 AM, Graeme Winter <[log in to unmask]<mailto:[log in to unmask]>> wrote: > Dear Randy > > However, when this noise is 0 all that happens is the next greatest source of ignorance steps up to the plate to cause us problems - for example, what exactly should the background be if the pixel values are all 0, with a couple of 1's? For what it's worth, I get extremely perplexed when I get background values of 0 and 1 on a pilatus detector. I've always believed (perhaps erroneously) that you should measure some background value , just so you can be somewhat certain that if there was a background, at least have some confidence in its estimate. Is it possible to have a background of zero, and if it is, how would you measure it and what is your confidence of that measurement? Again XDS, makes a recommendation for weakly exposed frames (http://strucbio.biologie.uni-konstanz.de/xdswiki/index.php/Low_dose_data), but it seems to be a workaround to a fundamental (?) question. F