Heatmap showing top 100 probe sets after k-means clustering (k = 20). Training data (n = 325) were clustered by expression value using k-means clustering (k = 20) for the top 100 probe sets identified by random forest classification variable importance. The first color side bar on the left indicates cluster number and the second indicates relative variable importance within the cluster (darker blue = greater importance). The top side bars indicate risk group (low, intermediate, and high from left to right) and relapse status (red = relapse; yellow = no relapse). Genes (probe sets) are indicated on the right axis. Genes highlighted in yellow represent the primary genes in the model (best in each cluster). Genes not highlighted represent alternates to primary genes in each cluster. Genes highlighted in pink represent genes excluded from the model because of probe set sequence ambiguity or status as a hypothetical protein.