ABSTRACT
In this chapter we discuss ways in which inferred networks can be tested for their validity. The tests we consider fall into two broad categories: direct tests of experimentally or computationally inferred networks via further experiments or via confrontation with other data sources-these sources include other networks, expression data, and functional dataand indirect tests of the predictions of the network model. An example of an indirect test is the usage of a protein interaction network to predict protein complexes, whose occurrence in vivo can be independently verified. There is necessarily some overlap between the examples discussed
here and those discussed in Chapters 3 and 4 on inferring the networks themselves, because it often occurs that the process of validating the predictions of a network model leads to refinement of the network model. This is what happens, for example, when different highthroughput datasets are combined to yield a more accurate network. Furthermore, predictions of one type of network model could well constitute another type of network model. For example, in the previous chapter we outlined how a protein interaction network could be used to predict a genetic interaction network. This prediction task could be viewed either as inference of the genetic network model or, if the predicted genetic network can be independently verified, as a test of the underlying protein interaction network model.
