Functional assessment of time course microarray data

Abstract

Motivation: Time-course microarray experiments study the progress of gene expression along time across one or several experimental conditions. Most developed analysis methods focus on the clustering or the differential expression analysis of genes and do not integrate functional information. The assessment of the functional aspects of time-course transcriptomics data requires the use of approaches that exploit the activation dynamics of the functional categories to where genes are annotated.

Methods: We present three novel methodologies for the functional assessment of time-course microarray data. i) maSigFun derives from the maSigPro method, a regression-based strategy to model time-dependent expression patterns and identify genes with differences across series. maSigFun fits a regression model for groups of genes labeled by a functional class and selects those categories which have a significant model. ii) PCA-maSigFun fits a PCA model of each functional class-defined expression matrix to extract orthogonal patterns of expression change, which are then assessed for their fit to a time-dependent regression model. iii) ASCA-functional uses the ASCA model to rank genes according to their correlation to principal time expression patterns and assess functional enrichment on a GSA fashion. We used simulated and experimental datasets to study these novel approaches. Results were compared to alternative methodologies.

Results: Synthetic and experimental data showed that the different methods are able to capture different aspects of the relationship between genes, functions and co-expression that are biologically meaningful. The methods should not be considered as competitive but they provide different insights into the molecular and functional dynamic events taking place within the biological system under study.

Figure 1

Schematic representation of the proposed methods. a) maSigFun fits a regression model for each gene expression submatrix defined by the genes annotated to a given functional class (FC.1 to 4 in scheme). Significant classes are obtained by the maSigPro method (FC.3). b) PCA-maSigFun obtains a PCA model for the gene expression submatrix defined as in maSigFun and extracts a number of components that collect non-random variation. Generally 0 (FC.1) to 2 (FC.2) components are extracted for each functional class. A regression model is then fitted to the scores vector of extracted components to select function-defined patterns with a significant association to time (FC.2 and FC.3). c) ASCA-functional applies ASCA-genes to identify principal patterns of variation associated with time and time × treatment experimental factors (PC1 to 3 in scheme). Genes are ranked by loading value in each PC, and GSA analysis is applied to each loading value-ordered gene list to identify a functionally related block of genes associated with the principal patterns of variation (FC.2 and FC.3).

Figure 2

Results of simulation study A with the maSigFun method. Changes in sensitivity with the percentage of co-expressed genes in the class at four values of the goodness of fit R2 of the regression models. Data points correspond to the mean value of 50 simulations. Confidence intervals were omitted due to their negligible size.

Figure 3

Results of simulation study B with the maSigFun method. Changes in sensitivity with the size of the class at three levels of percentage of changing genes (co-expression) in the class. One plot is provided for each level of the goodness of fit R2 of the regression models. Data points correspond to the mean value of 50 simulations. Confidence intervals were omitted due to their negligible size.

Figure 4

Results of simulation study A with the ASCA-functional method. Changes in sensitivity with the percentage of changing genes (co-expression) in the functional class. Data points correspond to the mean value of 50 simulations. Confidence intervals were omitted due to their negligible size.

Figure 5

Gene expression profiles for two significant representative GO categories obtained by maSigFun analysis of the Toxicogenomics dataset. a) Ferric iron binding category induced by high Bromobenzene and b) Bile acid transporter activity category repressed by high Bromobenzene. On the left panel, the median value of the functional class is plotted while the right panel shows the expression profiles of all genes annotated in the class. Treatments are labeled by color: pink HI, light blue ME, dark blue LO, green CO and red UT.

Figure 6

Score vs Loading analysis of PCA-maSigFun results on the Toxicogenomics dataset. a) Score profiles for three representative GO-components. b) Loading plot (gene contributions) for the same GO-components, genes labeled by their array ID. Blue lines indicate the threshold for significant contribution obtained by re-sampling (see methods).

Figure 7

Principal variation pattern of acute-phase response GO category in Toxicogenomics dataset analyzed by PCA-maSigFun. a) Scores plot reveals the profile of the GO-component. b) Loadings plot show gene contributions. Threshold for significant contribution are indicated by blue line. Names of positively correlated and negatively correlated significant contributing genes are indicated.

Figure 8

Principal variation pattern in the Toxicogenomics dataset. The pattern is captured by the first component of submodel b+ab (treatment + timextreatment) of ASCA-functional analysis. The plot shows the score values of this first component.

Figure 9

Principal variation pattern in the Potato Stress dataset. The pattern is captured by the first component of submodel b+ab (treatment + timextreatment) of ASCA-functional analysis. The plot shows the score values of this first component.