Re: very rare events: how low can you go?

From: Mario Roederer (Roederer@drmr.com)
Date: Tue Jan 30 2001 - 10:13:02 EST


Lynn,

I too encounter this same sort of problem--there seems to be some
sort of block in many people's minds that "1%" is some sort of
magical threshold, below which anything is deemed random.  I don't
know what sort of "fuzzy math" is behind this sort of
"undercounting", but it's pretty clear that the losers in this would
be the people who don't realize that there's lots of information in
even a single event.

In terms of statistics, the relevant value is the absolute number of
events that you are counting--the counting error on this is the
square root of this number.  Therefore, if you are analyzing a 1%
population, and collect 10,000 total events, your base error on the
frequency is 10% (i.e., 1% of 10,000 is 100; the square root of 100
is 10; 10/100 is 10%).  If you analyze the exact same cells, but
collect one million events instead, then your frequency error is now
1%.  (1% of 1M is 10,000; square root of 10,000 is 100, which is 1%
of 10,000).

Of course, this presumes that you don't have significant background
events.  I think many people are certain that "background" events are
1%, but this is, in words my 3-year old can hear, "bull hockey".
Background frequencies are entirely sample dependent, and can be
increased or reduced by any number of methodological steps.

That being said, my own experience with rare cells isn't as
substantial as others in the field.  Nonetheless, you can look at our
Nature Medicine paper in 1999 (Lee et al.) on identify
melanoma-specific T cells.  These were present at frequencies of
0.001 to 0.1% of PBMC.  Even at this low frequency, we were able to
phenotype them in detail, do cytokine analysis after stimulation, and
on one sample, do a V-beta repertoire serologically (using the
Immunotech VBeta kit).  In addition, we sorted the antigen specific T
cells (when they were present at the 0.01 to 0.1% level) for
stimulation and/or CTL assays--and it worked.

So to all those "one-percenters" in the world, tell them they're
missing out on more than 99% of life.

mr

(PS, I  am presuming, of course, that you are collecting more than
100 events per sample.  If you collect only 100 events total, then
those "naysayers" are correct.)



This archive was generated by hypermail 2b29 : Sun Jan 05 2003 - 19:01:04 EST