Graph analysis programs suitable for FACS

[Petr Kuzmic]

Ian A. York wrote:
> 
> I have FACS tracings (from a Becton-Dickinson FACScalibur machine, with
> CellQuest for the analysis) and I'd like to do some slightly more
> sophisticated analysis of them.  Here's my problem:  My cells are
> biphasic (the "High" and "Low" populations can't be distinguished by other
> antibodies); the curves of the two populations are overlapping; but I'd
> like to get some reasonably accurate stats on the two populations.

If you think that your histogram


                       ###                       
                     ########                    ##
                   #############               ######
                  ################            #########
                 ###################         ###########
                ##########################################
               #############################################
              ###############################################  
             ##################################################
           ######################################################
          #########################################################
        #############################################################
      ##################################################################


is an overlap of two NORMALLY distributed populations (or other such
well-defined populations) you can use any standard curve-fitting program
to deconvolute the two peaks.  I recommend the program SigmaPlot from
Jandel Scientific.  If the disctribution was Gaussian, in SigmaPlot you
would write the following transform file (replacing '...' with the
initial estimate relevant to your data):

-------------- 8< ------------- 8< --------------
[Parameters]
k1=... ; height of peak 1
m1=... ; center of peak 1
s1=... ; width  of peak 1
k2=... ; height of peak 2
m2=... ; center of peak 2
s2=... ; width  of peak 2

[Variables] ; these are your data
x=col(1) ; 'x' lives in column 1
y=col(2) ; 'y' lives in column 2

[Equations]
f= (k1/s1)*exp(-0.5*((x-m1)/s1)^2)+(k2/s2)*exp(-0.5*((x-m2)/s2)^2)
fit f to y
-------------- 8< ------------- 8< --------------

Good luck...

_____________________________________________________________
Petr Kuzmic Ph.D. * BioKin Ltd. * Madison, WI 53708-8336, USA
pkuzmic at biokin.com * http://www.biokin.com * 608.256.1269 fax