Analysis of un-replicated time-course microarray experiments

Since transcriptional control is the result of complex networks, analyzing dynamical states of gene expression is of paramount importance to detect the multivariate nature of biological mechanisms. Although hundreds of studies fully demonstrated the relevancy of microarrays in describing different physiological conditions, to reconstruct complex interaction pathways it is necessary to analyze the temporal evolution of transcriptional states. However, a robust experimental design for identifying differentially expressed genes over a temporal window would require large amounts of microarrays. Unfortunately, replicates for each time point and experimental condition are not always available, because of cost limitations and/or biological samples scarcity. In addition, common data analysis tools, like ANOVA, require replicates and disregard correlation structure among times. We present a method for the identification of differentially expressed genes in un-replicated time-course experiments. The procedure does not assume any model or distribution function, takes into account the correlation of data, and does not require sample replicates at the various time points, other than the presence of an initial time point for all analyzed conditions. The identification of differentially expressed genes as the result of a system perturbation is formally stated as a hypothesis testing problem in which a defined statistic is used to rank transcripts in order of evidence against the null hypothesis. Specifically, i) data are structured so that measurements are correlated in time, within the same biological condition; ii) the null hypothesis is formulated so that changes in expression levels at different time points are equivalent; iii) time point t0 represents the system before the perturbation. Therefore, modulated genes are detected testing the statistical significance of expression differences between physiological states at each time point, once corrected by the variability at t0, and given an empirical null distribution constructed using permutations. Statistical significance is assessed by the q-value. The method has been tested on time-course microarray experiments aimed at studying the temporal changes of gene expression in: i) skeletal muscle cells treated with a histone deacetylase inhibitor (Iezzi et al., Dev Cell; 2004) and ii) immature mouse dendritic cells (DC) exposed to larval and egg stages of S. mansoni (Trottein et al., J Immunol; 2004). Differentially expressed genes, identified using the proposed algorithm, have been compared with results obtained from ANOVA model and SAM paired test. The biological significance and soundness of selected transcripts was also verified using global functional profiling by means of OntoTools. Results demonstrate that this novel procedure allows the identification of biologically relevant genes using half of the replicates required by standard model-based approaches.