Fig. 3

Selection of the Clustering Algorithm and Number of Clusters using as Criterion the Clustering Silhouette. a Algorithm and number of clusters selection. The absolute value of Pearson correlation coefficient between gene expression profiles was used as gene similarity measure to perform our clustering analysis. The performance of the clustering algorithms hierarchical clustering (HCLUST in red triangles) and partition around medoids (PAM in blue squares) were compared for different number of clusters ranging from 4 to 20 using the clustering silhouette. The highest silhouette value was reached for the PAM algorithm with nine cluster (marked with an arrow). b Silhouette for PAM with nine clusters. The silhouette of a clustering measures both the inter and intra cluster similarities. The best clustering silhouette obtained with the PAM algorithm for nine clusters is shown. Each horizontal line represents a gene in a given cluster. A high positive value indicates a gene with a high intra cluster similarity and a low inter cluster similarity. Whereas a negative value indicates a gene with a low intra cluster similarity and a high inter cluster similarity. Genes belonging to the same cluster are represented with the same colour. For each cluster from one to nine, the number of genes and its average silhouette are specified