Re: best multicolor instrument?

From: Howard Shapiro (hms@shapirolab.com)
Date: Thu Oct 19 2000 - 18:01:30 EST


Rafael Nunez wrote:

I guess the multicolor mud is getting hot because of the off line
compensation or can we call it the software debate?
FlowJo has a feature for off line compensation (Mario is right!) as well as
FCSPress (Rick is right!) but this two softwares are not alone in this
feature (both are wrong!). The software from Partec, FlowMate has also this
feature for "to do compensation after acquisition and/or during analysis"
(it is good to have more options because of price). My educate guess is
that in a few more months we are going to see more off line compensation
features in the software market because of the competition. However, you do
not need off line compensation for doing multicolor studies. Am I wrong?

If compensation was set correctly when data were collected, there would not
be a need for off line compensation.

WinList (Verity) and FCS Express (DeNovo) also have facilities for off line
compensation, and the new software for the Beckman Coulter XL implements
full matrix compensation.

However, off line compensation requires that data taken on a log scale
(i.e., using log amps) be converted to a linear scale, compensated by
software in the linear domain, and then converted back to a log
scale.  Most flow cytometers don't collect data at more than 12 bits'
resolution; when these data are converted from log to linear and back, gaps
in the data develop.  WinList deals with the gaps by adding random numbers
to give the appearance most people are used to.

It is difficult to do hardware compensation for more than 4 or 5 colors;
the increasing complexity of the electronics increases the noise level,
decreasing dynamic range, and it is tricky to time the analog delays needed
for cross beam compensation.  Moreover, while it is possible to get
compensation for two or three colors right by twiddling knobs, this simply
means that the user is intuitively coming close to the solution of a 2 x 2
or 3 x 3 matrix equation, which is what is required for proper
compensation.  I suppose there may be autistic individuals (the so-called
"idiot savants" brought to popular attention years back in the movie "Rain
Man") who could intuitively compensate an 8- or 10-color experiment if
there were enough knobs available.  One could also, in theory, run the
single-color (and, occasionally, multicolor) controls needed to get the
coefficients of the matrix, solve the equation, and implement it via
hardware using multiplying digital-to-analog converters (DAC's).  Neither
of these solutions would get around the problems of electronic noise and
delay timing.  So what is left is software, typically implemented not in
the Windows or Mac computers used for data acquisition, but in digital
signal processing (DSP) chips, which are optimized for fast matrix
multiplication.

Cytomation and Becton-Dickinson have taken different approaches to the
problem in their latest designs.  Cytomation uses log amplifiers, and
digitizes the log signals to 16 bits, allowing conversion from log to
linear and back without degrading the data; the matrix equation is solved
by a fast DSP chip, allowing high speed sorting to be done on properly
compensated data.  The new B-D DiVa system uses high speed (14-bit, 10 MHz)
analog-to-digital converters to collect 32 points from each pulse; this
digital pulse processing yields an effective 19-bit signal, which can also
be converted from linear to log and back without degradation, and also
solves the matrix equation and compensates the data fast enough to allow
high speed sorting.  Data collected from either of these systems can also
be compensated off line without degradation.  While Beckman Coulter may
have an equivalent combination of hardware and software in the works for
its Altra sorter or a descendant, I haven't heard about it yet, so, for the
time being, the choice of a multicolor (8 or more colors) sorter seems to
be between B-D and Cytomation.

If the higher end Partec systems are digitizing to at least 16 bits, they
should produce data suitable for off line compensation.  The same would be
true for the B-D LSR, although I'm pretty sure it doesn't have 16 (or more)
bit conversion, which would tend to rule it out.

Eventually, everybody should be moving toward digital pulse processing;
this is much easier to do for relatively slow (a few thousand cells/sec)
analyzers than for high speed sorters, because one of the limiting factors
is the available range of high-speed, high-resolution converters.  At least
14 bits are needed to get a 4-decade dynamic range; B-D can get 32 slices
of its 3.2 usec pulses with a 10 MHz converter, but Cytomation would need a
40 MHz converter, because the pulses in the MoFlo are only 800 nsec
long.  Such a device has become available in the last 6 months, and I'd be
willing to bet that some of the first samples available went to
Cytomation.  Meanwhile, Luminex has been using 14-bit, 3 MHz converters and
a single inexpensive DSP chip in their production Model 100 bead analyzers,
and their entire circuit is considerably smaller, less complex, and
relatively expensive than what's in the sorters.  Similar hardware should
find its way into the next generation of benchtop cell analyzers, making it
easier both to compensate right the first time and to make adjustments
after the fact.

As I sang at the GLIIFCA Meeting in Motown last weekend-

The future of cytometry is a chip they call a DSP
Once we used log amps for dynamic range,
Now, the paradigm's about to change.
Where did I get the good advice to process pulses slice by slice?

You know, I heard it through the baseline;
Smaller signals will get through fine,
Yes, I heard it through the baseline, and the cost is apt to be benign
For the bottom line.

A 14- or more bit ADC gets you from ten volts to 1 mV.
Sampling 16 times as pulses pass gives your integral a touch of class;
You're better off with 32, but, in a pinch, even 8 will do.

I know, I heard it through the baseline...

Once you've learned to play with these new toys,
You can compensate for laser noise,
Measure the free dye in the stream, trigger on signals from any beam.
But you'll have CD's stacked in piles if you try to save raw data files-

It's so; I heard it through the baseline...

-Howard



This archive was generated by hypermail 2b29 : Sat Mar 10 2001 - 19:31:35 EST