The arrays were controlled for spatial and intensity dependent effects by visual inspections of the array images and of ratio/intensity (RI) plots. To retain saturated spots (with more than half of the pixels in saturation) we performed additional lower scans of the arrays and modified the saturated intensities according to the algorithm described in Lyng, Badiee et al (2004). As measures of the spot intensities we used the median of the foreground pixel intensities, and log-transformed them to obtain approximate normality. No further normalisation was done because this is taken care of in our model-based analysis described below.
Description
Hepatic gene expression was examined in abcb4 (-/-)mice (FVB.129P2-Abcb4tm1Bor/J) at 3, 6, 9 and 20 weeks after weaning, using spotted cDNA microarrays; FVB/NJ abcb4 (+/+) mice serving as controls.
Data processing
Our model is based on a log-linear mixed effect model according to Kerr et al (2000).The base2 logarithm of each of the measured fluorescent intensity was modelled as a sum of dye, gene and array effects plus effects of the interaction terms array*gene (i.e. spot effects), dye*gene and time*variety*gene (the time-dependent effect of the specific variety - here abcb4 (-/-) orcontrol - on each gene). The latter is the parameter of main concern, and we note that the difference here for a given gene between abcb4 (-/-) and control is the log2 ratio (normalised for experimental noise) in gene expression between the two types. Parameter estimates were obtained using the MicroArray ANOVA (MAANOVA) package by Wu et al (2002), for the statistical language R. We set a rather conservative threshold of significance of at least two-fold change and p<0.01, which corresponds to false discovery rates (FDR), i.e. the proportion of falsely claimed differentially expressed genes, of 2.6%, 2.5%, 2.7% and 2.2%, for the four time points, respectively, calculated in SAM (Tusher et al 2001).
The Gene Ontology (GO) based data mining tool High Throughput GoMiner (Zeeberg et al 2005) was used to find GO categories, i.e. molecular functions, biological processes, and cellular components, overrepresented by regulated genes.
