Re: Data Analysis Software

From: Thomas Delohery (t-delohery@ski.mskcc.org)
Date: Thu Mar 18 1999 - 19:28:17 EST


I must be missing something here.  I don't understand the point(s) being
debated or the issues being raised.  But I'll throw my two cents in anyway.

Several years ago I was evaluating cell cycle analysis programs;
ie, curve-fitting algorithms for analyzing histograms from cells with
fluorescent stained DNA.  I used BrdU incorporation as my Gold standard.
After analyzing numerous data files there was only one program that
consistently
gave values for percentage of cells in G1, S and G2/M that agreed with the BrdU
data.  The program always gave a slightly LOWER S-phase (and correspondingly,
slightly higher G1 and G2/M) than the BrdU incorporation data. In looking
at DNA
histograms of BrdU positive cells I noticed they usually showed a fairly
discreet
G2/M peak.  It then occurred to me that the curve-fitting program was giving a
BETTER measure of the percentages cell cycle compartments than the BrdU
data. BrdU
incorporation measures all cells in S-phase over the period of the BrdU pulse.
That includes the "dim" BrdU positives, with "apparently G1 DNA content"
that are
in the earliest stages of S-phase to cells that completed S-phase very soon
after
the BrdU was added.  These would be BrdU positive G2 cells.  So the fact
that the
curve-fitting program was giving values for percent S-phase below those
measured by
BrdU incorporation makes perfect sense.  DNA staining alone can't be
resolve late-G1
cells from early-S and late-S from G2.  That's why a GOOD curve-fitting
algorithm
will have an area of overlap at these transitions.

It's a little alarming to hear Adam Treister say:

>But for me, the role of data analysis software is to enable exploration
>and to test
>potential models, not only to churn existing models.  Mark's message lists
>valid uses
>for the capability. Clearly there are limitations, but this is flow, so
>what else is
>new?  The software is there only to calculate the numbers; it's up to the
>biologist
>to make sense of them.

One would hope that the software calculations are providing realistic
numbers so
biologists can make sense of them.  Esp. given the fact that I just ordered
3 copies
of FlowJo.  The last thing I need is a repeat of something like CellFit
where the
numbers calculated are heavily dependent on the model one chooses to fit
the data.
I wouldn't recommend you be too quick  to discard the old models while
you're exploring
and testing "potential " models.
td


--
==============================================================================
 Thomas M. Delohery                       |Internet: t-delohery@ski.mskcc.org
 Supervisor, Flow Cytometry Core Facility |   Phone: (212) 639-8729
 Memorial Sloan-Kettering Cancer Center   |   Fax: (212) 794-4019
 1275 York Ave. Box 98                    |
 New York, NY    10021                    |
============================================================================== 



This archive was generated by hypermail 2b29 : Wed Apr 03 2002 - 11:53:16 EST