We have discussed some statistical issues in the analysis of microarray time course experiments, touching on their design, the identification of genes of interest, clustering, and alignment. For the practitioner, we have tried to offer some ways of addressing these issues, particularly the second.

As will be apparent from our discussion, many, perhaps most, of the methods in the literature available for choosing or clustering genes in time course experiments have been devised and tested on the yeast and human cell-cycle datasets. There is room for much more research on the analysis of what we have called developmental time course data, especially their clustering.

A less obvious bias in our coverage is the fact that almost all of the methods we have discussed have been for data generated in an experimental setting, with mRNA from cell lines, tissue samples or experimental organisms such as whole Drosophila embryos, or tissue from inbred strains of mice or Arabidopsis plants. Recently microarrays have moved to the wider clinical setting, with microarray data now being collected on human subjects over time. Such longitudinal studies present novel analytical challenges, as subject-to-subject variation, even within the same treatment group, can be substantial. The methods we have reviewed here will not be appropriate in the clinical context without modifications, for example, by including fixed or random effects for subjects. This is an important area for future research, but we can refer to Storey et al. (39) for a promising start.

0 0

Post a comment