[Cytometry] Scaling problem with FACSDiva
Dave.Dunaway at nationwidechildrens.org
Thu Feb 6 09:30:31 EST 2014
As I said in my original response I understand that as Dr. Roederer put it what we are ultimately interested in is the numbers and I understand what you and he are saying here. That however does not address the point that I raised regarding how things look. You can minimize that but the reality is when it comes to presentation such as in a manuscript visualization does matter IMHO. If you accept that how do we react to data that is presented as positive with no negative to compare it to? Perhaps I misunderstood the original post but it seemed to say that in the case with a very bright positive signal that is smashed against the axis that its ok to bring it down even if that means you take the negative completely off scale in the opposite direction.
From: cytometry-bounces at lists.purdue.edu [mailto:cytometry-bounces at lists.purdue.edu] On Behalf Of Cris Bare
Sent: Thursday, February 06, 2014 4:40 AM
To: Purdue list
Subject: Re: [Cytometry] Scaling problem with FACSDiva
Well, Mario beat me to the reply, with his usual eloquence.
I will add an example, though.
Not too long ago I was teaching a colleague titrations using a FACSCanto II. Samples were single color and display plots were univariate histograms.
DiVa did a decent job of Biex transform and during acquisition I had clear positive and negative peaks.
When I brought the data into both Flowjo vX, and FCSExpress 4, the histograms suddenly displayed bizarre triple peak negative populations, and automatic scaling pushed them clearly into negative number space despite having zeros for the compensation matrix. I could reproduce the acquisition histograms by manually adjusting width basis (Flowjo) and log/lin transition points (FCSE). It seemed there was quite a lot of baseline restoring in this data, which I suspect was due to the cleanliness of both samples and system.
Regardless of how odd the histograms looked, when the stain index was calculated, the teaching point of the titration became clear. I wouldn't use those plots in any presentation, but the underlaying data said exactly what I wanted to hear.
On Wed, Feb 5, 2014 at 6:28 PM, Mario Roederer <roederer at drmr.com> wrote:
> Dave Dunaway wrote:
> > I find this to be dubious advice unless all you are interested in is
> crunching numbers.
> But, in fact, that's ALL we should be interested in. Graphs are for
> illustration purposes, for data exploration purposes, and for pretty
> pictures. We should not draw conclusions from graphs, we can only do
> so from numbers, properly controlled, properly analyzed, but numbers
> > The data may be there but if its not visibly demonstrative then your
> data is questionable when it comes to presentation.
> "Visibly demonstrative"? I'm not even sure what you mean by this.
> But it illustrates a basic mistake we all make -- we put too much
> emphasis on visualizations, on graphs. Data is questionable based on
> evaluations of its reliability and reproducibility (statistics). We
> can look at graphs to decide on what questions to ask of the data, but
> we must be cautious about drawing conclusions from graphs.
> I can show you examples of graphs which look great but are meaningless.
> And I can show you graphs of data that look questionable but are good.
> Cris's basic advice was correct, but brief. If your positives are
> within a half decade of the top end of the dynamic range, then reduce
> your PMT voltage (sensitivity). Inserting an ND filter is not
> necessarily an equivalent (or good) solution -- it accomplishes
> roughly the same thing but at a cost of increased noise from decreased photoelectron counts.
> Reducing the PMT voltage won't make negatives "invisible" -- it will
> only squash them near the zero value. Nothing is hidden.
> BTW, when you see your negative population go in to the "below-zero"
> measurement range of uncompensated parameters (ie towards the left
> edge on the transformed logicle scale), this is a consequence of the
> baseline restore in the DIVa electronics. It's a pain... but it's not
> typically a problem.
> Just make sure that you consider ALL events below the upper end of an
> unstained sample (wherever it ends up on your settings) to be equivalent.
> i.e., if your unstained cells go to 100, then EVERYTHING below 100,
> whether it's +10 or -300 is the same -- don't gate separately those
> events, even if they look like distinct populations in your graph.
> Because graphics can lie, but numbers always tell the truth.
> Cytometry mailing list
> Cytometry at lists.purdue.edu
> Search the list archive at http://tinyurl.com/cytometry
More information about the Cytometry