RE: MFI vs %Positive: myths deserve answers

I was surprised to find Mario's sketch to lack the lin or a log scale.
His example could actually also be used to argue against 1-D region as
the low par1 positive data also show a shift in par2. However what it
should tell us is that in order to understand our multicolour data on a
2-d level we should at least plot all fluorescence channels against all
scatter channels and al fluorescence channels against each other ungated
or colour gated to see correlations on a 2d basis. For the bog standard
3 parameter analysis that makes a column of log side scatter (RALS) vs
log fl1, fl2 and fl3, one of log forward scatter (FALS) vs log fl1, 2
and 3 and a set of FALS versus RALS as well as FL1 vs 2, FL1 vs 3 and
FL2 vs 3. Only if you have looked at all of these unrestricted data you
can have an idea about were to set regions. If someone presents me with
a histogram he might as well not waste the time on the flow cytometer as
he has clearly not understood the power of signal correlation which is
central to the technology. There is also a level of reassurance that a
bivariate displays can give to demonstrate the absence of a correlation.
And they are not any more expensive to generate from the computer.
Whatever cell population one looks at- there is always a subpopulation.
With the complexity of the data generated and considering the experience
level of most investigators these days it is impossible to judge the
quality of the data analysis if you are just presented with a histogram.
It more implies that someone wants to hide something or oversimplify. At
least the referees of an article should be given access to a
representative set of complete data files for data analysis QC.

Even if we can detect the presence of auto fluorescence we still have to
keep our brains in working order. It does not help to just be able to
subtract it as this might well increase our noise. It might be better to
change the fluorochrome to be further away from it or consider if it is
a cell volume related response. If, like in some cell cultures, it is an
expression of lipid oxidisation in the membrane as a response to
oxidative damage, we should consider to eliminate it as it is a problem
in our assay per se. The biggest danger to science it to assume that all
the odds are taken care of by the computer, so they do not need to
understand the process. In theory we can eliminate all these
autofluorescence artefacts, but as common wisdom tell us:

In theory there is no difference between theory and practice
- but in practice there is!

Happy flowing

Gerhard



-----Original Message-----
From: J. Paul Robinson [mailto:jpr@flowcyt.cyto.purdue.edu] 
Sent: 20 September 2004 21:13
To: cyto-inbox
Subject: Re: MFI vs %Positive: myths deserve answers


So, one of the real issues here is the rather traditional way 
we collect data without attempting to collect true spectral 
information. If we were able to collect relatively complete 
spectral scans of cells (which we can - because we just did 
it!) you would avoid a lot of this crap about what is 
autofluorescence and what is not! This is simply not an 
issue when you do a more complete spectral analysis. 
There are now numerous examples of this in the imaging 
world.	Further as Frank says, it's time to move to a more 
advanced analytical model - one that looks at a complex 
data sets for what they are - complex data sets.

The history of flow cytometry is that we increase the 
compexity of our biological experiments and try to leave the 
analysis at pretty much the same level as we did in 1980. 
We even still have people plotting graphs that say FL1 Vs 
FL2 and the like - something that is totally meaningless! 

Since I am going out of town for a few days, I guess I will 
put on an extra cooling fan for our server .....

Paul Robinson
Purdue

On 20 Sep 2004 at 13:28, facs_copy wrote:

> Dear colleagues, the beauty of this list is that anyone can venture an
> opinion and, you guessed it, here's mine. I'd like to comment on one 
> aspect of the MFI vs % +ve discussion; the use of 2D displays or 1D 
> histograms.
> 
> Nate suggests that, to delineate positive cells, we "use a 2-parameter
> plot and a polygon region - NEVER a histogram". Mario argues that
"this 
> is purely a visual artifact .. myth .. EXCEPT when there is a 
> relationship between the expression of the dim marker and the 
> measurement on the other axis" and we might just as well stick to a 1D

> histogram.
> 
> Indeed so. So far I almost agree with both, although "NEVER" might be 
> a
> bit strong. However, in my view, finding a measurement for the other 
> axis that helps the delineation is no rare exception. In particular, 
> when the property masking the specific fluorescence shift is 
> "autofluorescence" (whether or not this is truly "auto" or something 
> exogenous picked up after some time in the gut or some time in cell 
> culture), it is worth seeking a detector that sees this
autofluorescence 
> but not the specific fluorochrome. In fact a noted cytometrist (forget

> which) once suggested we set aside one detector specially for the 
> purpose, with optical filter presumably somewhere in the yellow
region. 
> To explain why this helps, and not possessing the Roederer eloquence,
I 
> rely on the accompanying cartoon graphic. Fig1 shows how such a 2D 
> approach can assist with a clean, manual delineation of positives and 
> negatives that would not be possible using only the 1D histogram.
> 
> Further, to consider Mario's scenario where cells of different
> autofluorescence may shuffle around in the histogram, making it appear

> that *all* the cells shift in fluorescence after staining, I direct
you 
> to Fig2. In my imaginary experiment I invoke 2 populations, randomly 
> called "T" and "B", having different levels of autofluorescence and 
> which may respond to staining for the devil's CD marker, CD666. The 
> result may be "One Possibility", where *everything* gets weakly
stained 
> and shifts in the the green fluorescence direction or "Another 
> Possibility", where just the "T" cells stain for CD666. There is
clearly 
> no way to extract the truth from the 1D histogram but the 2D display, 
> that also detects the autofluorescence differences between the 
> populations, shows what's going on.
> 
> Therefore, my advice to anyone enquiring is that we should *always*
> *try* to find a second parameter that adds information to the display.
> 
> To drift slightly off the topic, I would even go further and say that,
> if we had the computer graphic tools, we could add yet another
parameter 
> and *try* to delineate our populations in a 3D display. The extra 
> parameters may be physical properties or, as Mario, the champion of 
> "another parameter" may advocate, immunofluorecence. Unfortunately, 
> since we humans are dimensionally challenged, our world view stops at 
> 3D. Thereafter, if we want to use *all* 8 or 12 or 16 dimensions of
our 
> cytometric data, we can forget manual delineation and rely on computed

> cluster analysis.
> 
> Anyway, that's what I think.
> 
> Frank Battye.
> 
> | | << Frank Battye PhD
> \__/ <<<< The Walter & Eliza Hall Institute
> ------!!<<<<<< 1G Royal Parade, Parkville
> /!!\ <<<< Victoria 3050, Australia
> o !! \ << ph: 61_3_9345 2541, fax: 61_3_9347 0852
> 
> 


J.Paul Robinson, PhD		 PH:(765)4940757
Professor of Immunopharmacology
Professor of Biomedical Engineering  
Purdue University	   FAX:(765)4940517
EMAIL:jpr@flowcyt.cyto.purdue.edu
WEB: http://www.cyto.purdue.edu     

Have you seen our new HCS webpage? http://www.cyto.purdue.edu/hcs