Figure 1

Summary of the analysis performed for each data set of the MADMuscle database. The estimated correct (two yellow stars) data set "GSE1004_GPL91" from Judith Haslett et al. [8], which explores molecular profiles of dystrophin-deficient (DMD, blue color) and normal (black color) human muscle, is used as an example. For every analyzed matrix, each row represents a gene, and each column represents a sample. Each cell in the matrix corresponds to an expression level, with red for over-expression, green for under-expression, and black for gene expression close to the median (see the color scale). Heat maps resulting from hierarchical clustering (genes and samples) of raw data from GEO (A) and re-normalized data (B) are given. The heat map of the distinct clusters (from 0 to 8) identified by the stable k-means procedure (C) is also displayed. After hierarchical clustering (genes and samples), data from each identified cluster (D) are also represented by a heat-map, given a quality estimation and functionally annotated. Among them, two clusters, which discriminate DMD from normal muscle, are exemplified (clusters 1 and 5). Cluster 1, estimated as a good cluster (three yellow dots), corresponds to 1,107 genes over-expressed in DMD muscle (DMD+) and is enriched in genes coding for proteins of the extracellular matrix. On the contrary, cluster 5, estimated as a correct cluster (two yellow dots), corresponds to 896 genes under-expressed in the DMD muscle (DMD-) and is enriched in genes coding for proteins involved in translation.