I agree with Joseph on the value of increased resolution gained with higher resolution digitization of the analog measurement--especially for DNA measurements when you want to resolve small differences in the DNA content of tumor cells or normal cells, or X- or Y-chromosomes. All of the resolution gain can be lost, however, if the picture--the computer screen display--contains fewer elements to represent the data than are contained in the data range itself. That is, if you have 1024 channel data and these 1024 elements are mapped to a histogram that is 256 pixels wide, you have gained nothing. That is, a histogram is simply a vertical bar graph of fluorescence intensity on the x-axis versus the number of cells (events, if we want to be agnostic) on the y-axis. So in essence, you are building a more realistic mosaic of your data by using finer pieces of tile to represent what used to be infinitely variable differences among the cells. Frequency distributions (1 dimensional or 2 dimensional), also have an event sampling problem at higher resolutions. That is, as you describe the variation among the light scatter, fluorescence intensity, etc. of cells with greater precision, you must have larger numbers of events in order to resolve subpopulations easily. For example, on a 64 by 64 channel bivariate plot, you have 64-squared or 4096 points or 4096 different classifications based on 2 parameters. For data digitized at 1024 channel resolution, you have more than a million different classifications. Hence, smaller differences may give somewhat vague looking clusters until you have collected enough events that the resulting distribution approximates a good sampling of the 'real' population range. Image 8-parameter data where you can define some very small subpopulations. It is difficult to decide if the several hundred events (out of 2 millions cells collected) spread out on your 1024 by 1024 plot are significant. There are lots of good reasons, to collect higher resolution data. For one, you want to fool around with derived parameters, higher resolution data is always better (and even more the 10 bit resolution is better). If you have too few events, it is always easy to opt for a coarser display, but you can never get higher resolution displays from lower resolution data--smoothing aside. So, I would say collect at the highest resolution that you can. Storage is very, very cheap now. (Anyone want some 8 inch floppy disks?) And 10-bit data files are only twice as big as 8-bit data files. Having said that, I again agree that for many studies, the difference between higher and lower resolution data may be negligible. Dave ======================================== David M. Coder, Ph.D. Consultant in Cytometry email: d_coder@msn.com or dcoder1@hotmail.com tel./messages: 206-499-3446 ----- Original Message ----- From: "Joseph Webster" <J.Webster@centenary.usyd.edu.AU> To: cyto-inbox Sent: Monday, March 18, 2002 3:13 PM Subject: Re: 256 vs. 1024 channels > > Hi Calman et al > The vast majority of work here is in your category of "less exacting" > so we routinely use 256 channels unless there is some real need for > higher precision. > > The extra resolution does make a lot of sense for DNA; when I was > looking at tumour ploidy some years ago we only had 256 channels, and > it would have been much easier to distinguish those double peaks if > 1024 channels had been available. > Very sharp CFSE profiles might also benefit from higher resolution. > > Storage space is becoming less of an issue now, but remember that a 1024- > channel data file is four times bigger than the same data in 256 channels. > > Let's use our resources wisely, but don't toss out the infant with the > detergent.... > > Cheers, Joseph. > > At 13:57 18/3/2002, Calman Prussin wrote: > >Several years back when we were analyzing large numbers of events on G1 > >Power PC Macs, I started doing our data collections using 256 channels, > >rather than the typical 1,024. The speed of analysis was increased > >markedly with this change. > > > >Common sense tells me that there should be little if any difference in the > >data generated using either 256 or 1024 channels. Perhaps my histograms > >will look a little smoother? > > > >Question: What real advantage is there to using 1024 rather than 256 > >channels? Does it add significantly to the precision of DNA, CFSE or CBA > >assays? For less exacting applications (phenotyping, intracellular > >cytokines) is there any real benefit? > >Thanks, Calman > > -- > Joseph Webster, Flow Cytometry Facility > Centenary Institute, Sydney, AUSTRALIA. > Phone +61-2-9565-6110 > > >
This archive was generated by hypermail 2b29 : Wed Apr 03 2002 - 11:59:29 EST