In order to predict more complex RNA-to-RNA relationships in the siRNA-treated A375 cells, including upstream regulators of the co-expressed clusters described above, and the putative direction of RNA-to-RNA relationships. Reassuringly, 226 of the 327 combined children of these three E2F transcription factors have E2F binding sites in their promoters, a significantly greater proportion than would be expected due to chance. As well as identifying hubs, Bayesian gene networks also identify clusters of co-expressed RNAs, which are downstream of the same hub. Identifying these clusters may be seen as a more conservative use of this network method than identifying directional edges, and is the primary use made of Bayesian gene networks in this paper. Reassuringly, every one of the 200 clusters identified by the hierarchical clustering method above had at least 70% of their members included among clusters identified by the gene network method. We wished to determine whether the Bayesian gene network hubs and clusters identified from the A375 microarray data were associated with prognosis. Therefore, we used the ��Survival�� package in R to generate Cox proportional Hazards models to estimate the association between the abundance of RNAs in tumours and the survival of melanoma patients. Two survival models were generated: based on gene expression in metastatic melanomas using an Affymetrix microarray dataset and based on gene expression in primary melanomas using an Agilent microarray dataset, which was mapped to Affymetrix probe IDs using Entrez gene ID annotations. We then used this melanoma microarray survival information to assess whether gene network hubs and clusters were significantly associated with patient survival. Firstly, to establish a baseline, we considered whether the abundance of RNAs that encoded AZ 960 proteins with particular classes of functional annotation were significantly associated with patient survival. We hypothesised that RNAs Screening Libraries clinical trial encoding the types of proteins that perform important oncogenic functions may be more strongly associated with the survival of patients than the abundance of RNAs that encode proteins that do not play known roles in cancer. For both primary tumours and metastatic tumours, no one functional category was clearly more or less associated with patient survival than all RNAs taken together. This analysis was repeated for all Bayesian gene network hubs with $50 downstream children but again it did not identify any particular functional category with strong patient survival associations. We then repeated this analysis focussing on hubs with children that encoded proteins of common function. We used the GATHER web tool to identify hubs with children significantly enriched for GO paths.