Application of cluster analysis of temporal gene expression data to panel data

Abstract

The objective of this work was to determine the best alternative for the formation of homogeneous groups of gene expression series among the hierarchical clustering (Ward) and optimization (Tocher) methods, and to perform predictions regarding the gene expression of these series from a small number of temporal observations. The data used refer to the expression of genes that act on cell cycle of Saccharomyces cerevisiae, and corresponded to 114 gene expression series, with ten-fold-change values (expression measure) each, over time (0, 15, 30, 45, 60, 75, 90, 105, 120, and 135 min). The parameter estimates of autoregressive models AR(p) were previously adjusted to individual series (from each gene) of microarray time series data and used as variables in the clustering process. Gene expression predictions were made within each formed group from the adjustments in AR(p) model for panel data. The Ward’s method was the more suited for the formation of gene groups with homogeneous series. Once these groups are obtained, it is possible to adjust the model AR(2) for panel-data, and successfully predict gene expression at a future time (135 min) from a small number of temporal observations (the nine other fold-change values).