[Cytometry] Proper usage of CV in data analysis

Mario Roederer roederer at drmr.com
Fri May 13 15:45:03 EDT 2011


CV is a fine assessment of variability of fluorescence (most programs compute log-CV for log distributions).  How you interpret the CV requires the same caveats than how you would interpret mean, median, IQR, or any other summary statistic.

For clonality of transfected populations, however, the answer is pretty much "absolutely not."  We used to routinely demonstrate that you could clone a stable transfectant, then sort the highest or lowest expressing populations, and in a vast majority of clones, the sorted population would remain in its new "higher" or "lower" position.  This indicated epigenetic changes (manifest at transcription levels) leading to highly disparate expression from a known clone.  In a few cases, integration was in a highly regulated site, and no matter what you sorted, the sorted populations would recapitulate the original distribution.  But that was the exception rather than the rule.

mr

On May 13, 2011, at 4:36 AM, Vinko Tosevski wrote:

> Dear all,
> 
> The other day a colleague asked me if there's a way to assess the
> clonality of her transfected cell population based on the width of the
> peak she's seeing in the cytometer. While thinking about this, I came
> up with some questions regarding the use of coefficient of variation
> in flow cytometry for assessing the dispersion of the measured
> population. I would appreciate if you could expand my perspective on
> these topics...
> 
> Can I draw some conclusions about clonality of the population by just
> looking at the width of the histogram peak? If yes, how and what's the
> meaning of it? What do I compare it with? I assume comparing it to the
> width of untransfected cells makes no sense since those ones are not
> carrying the "trait" that I am looking at to begin with (not
> expressing fluorochrome - tdTomato, by the way). How does one approach
> this?
> 
> Looking at the peak of untransfected cells made me ask myself another
> question - are CV values properly used in addressing the dispersion of
> the measured fluorescent populations? I mean, those are not normally
> distributed populations (they are log-normal at best) so use of CV
> would be inappropriate since assumptions of normality are not met. I
> guess robust CV would be the way to go, or...? But, in addition, does
> any type of CV makes sense?? If I remember my textbooks correctly, CV
> as a measure of dispersion only makes sense if one is dealing with
> measurements of a ratio scale (in sense of statistics). Could
> fluorescent measurements be actually seen as ratio scale? To me, it
> sounds more like interval scale, but am I right, then, to conclude
> that CV should not be used at all?!? At the moment I suggested her to
> use interquartile range as a measurement, but I am really looking
> forward to hearing some thoughts on this.
> 
> I appreciate your input!
> 
> Best,
> Vinko
> 
> 
> --
> Vinko Tosevski
> Institute of Experimental Immunology
> University of Zurich, Irchel campus
> Building 44, Room J42
> Winterthurerstrasse 190
> CH-8057 Zurich
> Tel:  +41 44 6353706
> Fax: +41 44 6356883
> tosevski at immunology.uzh.ch
> _______________________________________________
> Cytometry mailing list
> Cytometry at lists.purdue.edu
> https://lists.purdue.edu/mailman/listinfo/cytometry
> Search the list archive at  http://tinyurl.com/cytometry




More information about the Cytometry mailing list