Biology Reference
In-Depth Information
about the magnitude of parameter values, but even more about whether compo-
nents interact at all (both under the conditions studied and in living organisms).
To sketch what I will argue, fundamental features of living systems are crucial
to how we can deal with these data errors, and these features require recognizing
the developmental and evolutionary natures of these systems.
The new use of 'gene chips' presents an array of new possibilities in the
massive volume of data produced. Unfortunately, just when we would seem to
need more accurate data rather than less, they also appear to present us with
much higher error rates. These DNA microarrays use miniscule amounts of
different DNA sequences ('probes') to detect RNAs that may be involved in
producing active proteins. Chips with tens of thousands of distinct probes are
common. The latest and largest number is nearly 400 000 on a single chip. They
are used not only for broad censuses of activity, but also for more targeted ones
such as identifying interactions in specific metabolic pathways or disease states.
The targeting is as simple as the choice of what to spot in the array.
6
They can
either detect or compare activity patterns using several different protocols, but
so far in a boolean ('yes/no') rather than quantitative manner.
This new technology allows an enormous reduction of labor and coordinated
detection of simultaneous activity patterns involving multiple genes or proteins in
a cell, something that would have been impossible two decades ago. Gene chips
also increase the uniformity of assay procedures. The changes they have provided
are not unlike the move from 'single-unit' recording in the neurophysiology of
the late 1950s-1970s to the localization of massive changes in activity patterns
by brain regions possible with functional nuclear magnetic resonance (fNMR)
beginning in the 1980s. This transition produced not only new kinds of data,
but also inevitably new orientations in theory and is probably responsible for
the emergence of the new discipline of neuropsychology, which draws heavily
on the sort of molar data produced by fNMR. The change in orientation and
questions that could be addressed was enormous in both cases (though for gene
chips, still early in its course), but so also in each, the increase in breadth of
information was accompanied by a loss in its specific local quality. As a result,
in both cases, we have the continuation of two technologies rather than the
replacement of one by another.
Because the reduction in data quality with gene chips is substantial, its accu-
racy is a key issue. Deane et al. (2002) conducted an evaluation by compar-
ing the results of these 'high-throughput' methods with a database of already
known interacting and noninteracting proteins, producing the expression profile
reliability (EPR) index. A second method (PVM) involved determining how
likely the individual interactions were by asking whether they had paralogs
that interacted. The second method picked up only 40% of known interactions,
6
Wikipedia entry on 'DNA arrays', accessed 23 June 2006.
