between mouse var iation. False positive rates were estimated using p values that were calculated by permuting model residuals. Two types of multiple test corrections were performed. The p values were adjusted using the Sidak step down approach, and the Benjamini and Hochberg method. The qvalue software package was used to esti mate the number of genes that do not have significant between mouse transcript variation, ��0. To separately assess significance of between cage and within cage var iation, the following model was used, Each yikg is written as the sum of the average transcript abundance for that gene, ug, a cage specific effect, cig, a mouse within cage term, dj g, and a within mouse term, wikg. The Pritchard et al. data were revised to correct a processing error as previously reported.
For comparative purposes, we applied the same tests for significance of between mouse variation described above to the corrected data. Coexpression network analysis Variable genes were analysed separately for each tissue using coexpression networks. Every pair of genes was given a weighted connection, rs2, equal to the square of their correlation coefficient across all samples. Transcript abundance profiles were hierarchically clus tered and modules were obtained by a dynamic dendro gram cutting method Dacomitinib and subsequent module merge procedure. We only retained modules with more than 25 members. Modules are referenced by their tis sue of origin and by a colour index. For each module, the first principal component was computed to give a representative profile, referred to as the module eigengene.
We determined the sign of the module eigengene to be positively correlated with the majority of genes in the module and refer to this majority as the positively correlated module genes. The complementary genes are referred to as the nega tively correlated module genes. Module eigengenes were scaled to match the median variance over all genes in the module. For each gene, we computed the intraclass correlation coefficient, c �� sb2 as a measure of the relative contribution of the between mouse variance component. We decom posed each gene profile into a between mouse profile and a within mouse profile. The between mouse pro file averages the two samples within each mouse and the within mouse profile is the difference between sample 1 and the average value for that mouse.
To measure similarity of between and within mouse pro files, we computed Pearson correlation coefficients, rb and rw, for between mouse and within mouse profiles. When assessing significance of similarity of correlation among eigengenes, we applied a Fisher transforma tion with sample size n 11 and n 12. For significance a 0. 05, this required |rb| 0. 66 and |rw| 0. 64. Gene set enrichment Each module of the coexpression networks was tested for enrichment within the Gene Ontology gene sets and the Kyoto Encyclopaedia of Genes and Genomes pathway gene sets. The universe was defined as the set of variable gen