A simple but powerful tool for extracting temporal patterns is found in contrasts: linear combinations of gene expression measurements over time. Contrasts usually but not always have their coefficients summing to zero. An example of the use of contrasts can be seen in Lonnstedt et al. (49) where samples were taken from cells at 0.5, 1, 4, and 24 h after stimulation with a growth factor. Genes were regarded as early responders if they had large values of < c, E >= ctEt where ct = (t - 24.5)2 and Et is the gene expression value at time t, while those having large values when ct = t2 were termed late responding genes. Smyth (29) used contrasts in the univariate linear model setting, and derived a partly moderated F-statistic for testing whether there is any change in gene expression levels over time. This approach assumes the samples are independent, and so would be appropriate for cross-sectional data. Fleury et al. (50) described a valuable multi-criterion optimization method called Pareto front analysis, for ranking and selecting genes of interest. In their paper they made use of contrasts to select genes with many predefined patterns. In essence, Pareto fronts and their variants (50-52) seek to identify genes with large values for all of a set of competing contrasts of interest.

0 0

Post a comment