Accuracy of Microarrays

Aubrey de Grey ag24 at
Wed Jul 26 09:25:41 EST 2000

Florian Muller wrote:

> The experimental variation of an RT reaction (a key step involved in
> all microarrays) on the "signal" of a specific gene is between 40 and
> 60% in the hands of a professional. Statistical analyses rarely take
> this into account and you have people publishing 2 fold increases as
> "statistically highly significant". As a rule of thumb I would dismiss
> anything that is not at least 3 fold increased as a fluke or as
> unproven.

This completely true statement is beginning to gain acceptance, due in
large part to a characteristically trenchant expose by the inestimable
Rich Miller at the recent AGE meeting, in which he pointed out that the
number of false positives expected from these experiments, given these
vast variations, is of the same order as the number of actual positives
being reported.  A defence that some have offered is "OK, big changes
in an individual gene may not mean much, but when quite a lot of genes
involved in the same process are coordinately changed the result is a
lot more meaningful".  Unfortunately, this too is an extremely fragile
assumption, because it suffers from the "multiple hypothesis" problem:
one has to consider the likelihood of those genes being coordinately
changed in expression levels in the context of all the *other* sets of
genes that are *not* coordinately altered even though they are just as
related in function as the set that caught the microarrayer's eye.  It
is very difficult (though not impossible) to do statistics on such data
in such a way as to determine whether the observed result really is
statistically significant.  I'm optimistic that the recognition of this
problem will give rise to better analysis of such data in the future,
but it may be a while, because the fundamental limitation is that so few
animals are being sampled that it is almost impossible to get results
that reach significance.  Hence, longer and more expensive experiments
are the only way out.

Aubrey de Grey

More information about the Ageing mailing list