Multivariate Cox proportional hazards analysis was performed in SAS v9. 0 to estimate the hazard HDAC inhibitor Sigma ratio associated with cluster expression in the three groups after controlling for stand ard clinical predictors. Chi Square tests were used to examine correlations between cluster groups, individual genes, and tumor selleck catalog subtype. The pathway was built de novo based on information from KEGG, BioCarta, and a review kinase inhibitor AG 013736 by Yarden and Silowkoski with a focus on the RAS MEK and PI3K AKT components. Results Comparison of the GWAS catalog and Drugbank shows GWAS only detects a very small fraction of existing drug targets We examined the relationship between genes in the GWAS catalog and drug target genes in Drugbank. The GWAS catalog is a comprehensive collection of results from published GWAS studies on a wide variety of disease and other traits such as height. Drugbank is a database that combines detailed drug data with comprehensive drug target information. We compiled a list of disease related traits in the GWAS catalog and extracted the reported genes for each of them. The disease list includes a number of cancers, a variety of complex trait diseases, and disease predisposition traits such as obesity and hypertension. We then found the drugs used in treat ment of each of these traits in Drugbank, and extracted the drug target genes for each drug. Thus, for each trait, we have a list of GWAS reported genes and a list of drug targets. For the 88 GWAS diseases that have drugs in Drugbank, there are on average 29. 2 GWAS reported genes and 24. 0 drug targets for 19. 9 drugs. There are a total 23 instances of GWAS genes that are also drug targets for the same disease. Three of these genes are each drug targets for two different diseases, so that only 20 of the 856 drug target genes have been dis covered in GWA studies of the corresponding traits. This is slightly larger than the overlap of approximately 5 from a completely random model, but is a very low number considering that altered activity of most drug target genes will influence the disease phenotype. Possible data related reasons for low overlap One possible cause of lower overlap is that in Drugbank, some drug targets do not have a known mechanism and are probably predicted targets based on sequence simi larity to other verified drug targets, and thus may be incorrect. Thus, the number of GWASdrug target matches missed as a consequence of misidentification of candidate genes appears very small. A third data related factor is cover age by the tag SNPs on the microarrays used in GWAS studies. If there is no tag SNP in linkage disequilibrium with the underlying variant involved in a disease mechanism, that contribution to the trait will not be detected. A study of 160 non GWAS derived candidate genes for blood pressure concluded that only half were adequately covered with tag SNPs on a 500K array, suggesting this is a significant factor. But overall, data considerations do not qualitatively change the picture of very low GWAS genedrug target overlap.