FBPA clusters (-)-Nutlin-3 showed more noise than STEM clus ters, because all 238 genes were clustered. However, there appeared to be a general mapping between STEM and FBPA clusters. STEM Clusters 1, 4, and 6 mapped well to FBPA Cluster 1. STEM Cluster 2 mapped to FBPA Clusters 1 and 3. STEM Cluster 3 mapped partially to FBPA Clusters 1 and 2. FBPA Cluster 4, however, did not match any of the STEM clusters. Also, genes showing down regulation, repre sented in STEM Cluster 5, were included in FBPA Clus ters 1 and 2. Because the features selected for clustering did not emphasize magnitude of expression but rather rates of change, the down regulated genes did not cluster separately in FBPA. Interestingly, all significant STEM clusters showed some degree of mapping to the largest FBPA cluster, Cluster 1.

Clustering gene expression in the bystander response In order to compare the two clustering methods on a related cellular response, we applied STEM and FBPA to gene expression curves after bystander exposure to radiation. We discuss the results of clustering bystander responding genes using the STEM platform first. We selected the results from c 3 and m 100 for analysis of bystander gene expression. Again, results were rela tively consistent across input parameters. These para meters resulted in significant clustering of 160 out of the 238 cases. Figure 5 shows the gene expres sion profiles for the most significant clusters, 6 out of 100 possible clusters. The number of genes included in each cluster was again relatively uniform, ranging from 8 genes in Cluster 6 to 39 genes in Cluster 1.

Although the results visually showed good cluster tight ness, we noted that Clusters 2, 3, 5 and 6 looked rela tively similar, suggesting that these clusters represented subdivisions of a larger cluster, limiting the usefulness of the results, despite the use of 100 distinct profiles. Addi tional file 4 lists clustered genes from the application of STEM to the bystander gene response. The expression curves of the 238 genes in bystander cells were also clustered using FBPA. Again, to deter mine the optimal number of clusters, we used the gap statistic. We examined k 3 and 5, which both showed near zero inequalities. Average homogeneity was found to be 2. 376 and average silhouette was 0. 372 for k 5. For k 3, average homogeneity was 2. 950 and average silhouette, 0. 489.

Because reasonable structure and good tightness were found with k 5, we chose to present this clustering. The Rand index to the manually curated clustering was 0. 745, indicating high similarity equivalent to that of STEM. Additional file 5 lists clustered genes from the application of FBPA to the bystander gene response. The FBPA clusters are shown in Figure 6. The within method metrics indicate that GSK-3 Clusters 2 and 5 showed homogeneity and Clusters 3 and 5 showed good separa tion in terms of average silhouette.