Search the CONFOCAL archive at
http://listserv.acsu.buffalo.edu/cgi-bin/wa?S1=confocal

>Search the CONFOCAL archive at
>http://listserv.acsu.buffalo.edu/cgi-bin/wa?S1=confocal
>
>I have also done a real experiment with green, yellow, and cyan fluorescent
>proteins simultaneously in a mixed population of cells on the meta system
>and had no difficulty unmixing the spectra. I have also seen clear
>separation with a variety of other closely overlapping spectra. It does
>work.
>
>Regards,
>
>Kirk J. Czymmek, Ph.D.
>Associate Professor
>Department of Biological Sciences
>University of Delaware
>Newark, DE 19716
>(302) 831-3450

Hi Kirk,

Survivors of the UBC live-cell course know that "If it isn't
diffraction, it's statistics".  In the case of linear un-mixing, its
the latter.

For decades, linear unmixing has been applied with great success to
energy-dispersive x-ray spectra (where it is called deconvolution).
In this case too, the position and shape of the spectral peaks is
assumed to  be known before-hand.

X-ray spectra often have 1,000,000 in a peak, (divided into 10 - 20
channels) but much lower counts are used for x-ray mapping. Lower
counts boosts the noise level in each pixel, but having many pixels
consisting of the same composition gives "the eye" the illusion of
better data.

As several have attested, it can also be applied to signals from the
META detector.  However, it might be nice to hear how people "knew"
that it had "worked." Did they just look at the results on the
screen? Did they make all the 9 images for FRET or just look at some
final output? How did they know ahead of time which certain cells
"should" show one color over another? (i.e., maybe one half of the
pair just faded faster). How did they balance the stain
levels/laser-power, pinhole size so that all dyes present were
emitting detectable signal at about the same level? (Keep in mind
that the QE of the PMTs drop precipitously at longer wavelengths.)

Success in unmixing depends crucially on signal levels. Even with 100
detected photons/pixel, the statistical noise in the output of each
(say) 4 channels can be tremendous (about 20% likely error), so it is
hard to see how much careful fitting can be done over 4 channels.
This doesn't count the photons lost because they strike the glass
between the micro-PMTs.

If Zeiss "deconvolves" the data spatially (actually, 3D filtering)
before deconvolving in the spectral regime, it will greatly improve
the S/N.

All this means that if your specimen is bomb-proof (fixed and in
antifade: characteristics that may effect spectral properties but
permit 1,000,s of photons pixel?), unmixing is undoubtedly very
useful. This may be less true on living cells, where damage will be
greater (no anti-fade) and more "important" (not just fading but
death!), and where one may need to take several images to see some
change over time. Even ignoring the resolution loss (and hence signal
loss) due to "specimen-induced defocusing", these low signal levels
make it hard to see "Rayleigh criterion" resolution even if you count
100's of photons/pixel.

It would be really nice to know even a guess of how many
photons/pixel were detected in the studies cited.

Cheers,

Jim P.
--
               ****************************************
Prof. James B. Pawley,                             Ph.  608-263-3147
Room 223, Zoology Research Building,               FAX  608-265-5315
1117 Johnson Ave., Madison, WI, 53706  [log in to unmask]
"A scientist is not one who can answer questions
but one who can question answers."
                Theodore Schick Jr., Skeptical Enquirer, 21-2:39