3% to Cluster 5. Comparing the bystander FBPA clusters to STEM clusters, STEM Cluster 1 mapped well to FBPA method Cluster 2. STEM Clusters 2, 3, and 5 mapped relatively well to FBPA Cluster 1. As noted above, many of the gene expression curves assigned to STEM Clusters 2, 3, 5 and 6 showed a generally similar pattern. STEM Cluster 6, however, mapped most closely to FBPA Cluster 2. STEM Cluster 4 mapped partially to FBPA Clusters 2 and 4, while FBPA Clusters 3 and 5 did not match any of the STEM clusters well. Between Method Agreement After performing clustering on the microarray and qRT PCR data using the STEM software and the FBPA approach, we used the Rand index to compare the agreement of methods. The Rand index table indicates this was generally good across clusterings.

We note higher consistency between FBPA clusterings of the data than STEM clusterings of the data in both irradiated and bystander con ditions. Both the STEM and FBPA methods showed lower agreement with the manually curated standard for qRT PCR data than for microarray data as shown in the first row of Table 1, but the STEM clustering performed noticeably more poorly. As all clustering methods indicated relatively good clus tering agreements, we next examined the biological enrichment of individual clusters to explore the useful ness of the information generated by clustering genes by patterns. Network and ontology analysis for direct irradiation gene response We next analyzed individual clusters using biology based approaches that facilitate understanding biologi cally relevant responses.

The first approach was an ontology based analysis using the PANTHER database. We first considered STEM clustering of the irradiation gene response. As mentioned previously, STEM clustering provided six significant clusters with relatively uniform cardinality. We applied gene ontology methods using the PANTHER web based tool to assess the biological relevance of these six clus ters. We started by mapping genes in each cluster to functional Dacomitinib and pathway annotations in PANTHER. This step maps gene identifiers to annotations in the PANTHER database and is important because of redun dancy of biological annotations in databases, which may affect the outcome of analyses. We found that coverage of mapping in the six clusters was randomly spread from 67% in the largest cluster, Cluster 1, to 93% mapped genes in Cluster 2. Surprisingly, gene ontology enrichment showed that only Cluster 3 was significantly enriched for biological processes, which spanned diverse functions from apoptosis to cell signal ing and proliferation. Minimal biological struc ture was apparent in the other clusters.