Re: Mean Fluorescence Channel

From: NOVO,DAVID JONATHAN (dnovo@ucla.edu)
Date: Wed May 10 2000 - 17:21:58 EST


Hi Olindo,

	It depends what you mean by linearizing your data. If you just
mean changing the axes on the plots so that the scale is linear instead of
log then no - the numbers that you get from that will not reflect your
underlying data.

	It is possible to mathematically transform your data into
what it theoretically would have been before the log amplification had
been applied. Then you could perform a mean, median etc and get something
meaningful. HOWEVER, there are many pitfalls to this approach which makes
it a pretty lousy idea.

	Pitfall 1: Log Amps are not perfect. So, the number that you get
after converting to linear is not necessarily the same number that was fed
into
the log amp. Now, log amps these days are pretty good, and it is possible
to calibrate your log amp. I figure that if you had innacurate numbers
after the log transform, due to a lousy log amp,  then your data doesn't
represent your cells in the
log world anyhow. So - with careful calibration I think this source of
error may be minimized.

	Pitfall 2: This is the far more important source of error. On a 10
bit scale (i.e. 0-1023 channels) that is common for most of the cytometers
our there, you lose a lot of information when the data is changed from the
analog -
log amplified data to digital.	Its probably not worth it to get into all
the nuts and bolts of the whole thing but you can check out Practical Flow
Cytometry pp 163-170 for a discussion on the issue. Most programs that
allow you perform offline compensation (i.e. compensation after
acquisition) have to perform this calculation and as a result end up
adding random numbers into their data to restore the plots to have nice
fuzzy cloud populations instead of populations of straight lines that
actually result from the math due to rounding errors.

	So, the moral of the story that while its not impossible to go
from log acquired data to linear, its not the easiest thing in the world,
and 99.876% of the time is not worth the headache.

-Dave


*************************************************************
David Novo
Dept. of Physiology
UCLA
CHS C8-134
(310) 206-8160
*************************************************************

On Wed, 10 May 2000, Olindo Assis wrote:

>
> Dear All,
> I think I was not clear enough since some of you sent me questions
> concerning my issue. I will explain that in another words.
> What I want to know is whether or not I can analyze data using linear scale
> even when I have acquired them using log scale for fluorescence acquisition
> mode. In another words, can I linearize data after acquisition or this is
> not an acceptable toll. Actually I thought that I always need to use log
> scale on acquisition mode for fluorescence.
> Thank you for your attention,
> Olindo
>
>
>



This archive was generated by hypermail 2b29 : Sat Mar 10 2001 - 19:31:19 EST