Biology Reference
In-Depth Information
protein structure information or evolutionary signatures from the large
number of sequenced genomes. 4 In order to assemble a comprehensive
and reliable protein network, experimental interaction data still remain
essential. This, however, does not mean that computational predictions
would have no value per se — indeed, it is often advisable to include them,
for example, to augment and filter experimental data with predicted inter-
actions, much like experimental data can be filtered with previous knowl-
edge and with interactions determined in other organisms.
When working with experimental protein-protein interaction data, it
is important to distinguish between small-scale experiments that have
been individually reported in scientific publications and high-throughput
experiments that aim to describe a large number of protein-protein inter-
actions in a single dataset. Both approaches have their strengths and
weaknesses. Small-scale experiments are usually more reliable, since they
are often based on a combination of experimental techniques, have been
subjected to the scrutiny of peer review, and may have been reproduced
independently in other labs over time. However, they can be difficult to
assess in bulk unless curators have extracted them manually from the sci-
entific literature. Moreover, they usually contain a significant reporting
bias: negative results (noninteractors) are often not reported, and even
the positive results are sometimes focused on specific areas of interest or
have a tendency to confirm previous findings.
With high-throughput experiments, the situation is often the oppo-
site. Reporting bias is usually not much of a problem, since a large set of
protein interactions is assessed simultaneously. Similarly, the data are usu-
ally accessible quite easily in electronic format, and negative results (non-
interactors) are normally part of the reported data. On the downside, the
overall quality of high-throughput results often cannot match that of tra-
ditional, small-scale laboratory experiments. The rates of false-positive
and false-negative interactions are often considerably higher. This is usu-
ally inferred not only from observing unusually low reproducibility
between independent high-throughput datasets that aim to measure the
same interactions, but also from deviations of these data from previously
accepted reference data. 5-8
Numerous different experimental approaches can be used to assem-
ble the data for a protein-protein interaction network. These include
Search WWH ::




Custom Search