These gene lists were uploaded into the IPA software where a core analysis was completed to determine the association of each gene with various biological functions or network Pazopanib VEGFR pathways. IPA comparison analyses were used to reveal whether or not cells expres sing WT or KR PR upregulated functionally distinct path ways. Analyses were scored based on significance and the threshold for a gene list to be significantly involved in a particular biological function was P 0. 05 1. 30. Identification of PR expression metagenes Metagene analysis was conducted using gene expression microarray data from cell lines constitutively expressing empty vector, WT PR, or K388R PR, and treated with either vehicle or R5020. A strategy of identifying meta genes within each sample was employed using non nega tive matrix factorization.

This strategy facilitated identification of metagenes and application to other data sets. To limit the study to genes under high variance and to limit the number of probes used in calculating the metagene fit, probes were considered for metagene analy sis based on the interquartile range of the probe being in the upper 80th percentile. Inhibitors,Modulators,Libraries The optimum rank of the data was calculated as eight, therefore, eight meta genes are present in the data. Three of these metagenes were either highly expressed in all samples, or expressed in no samples, indicating that they are likely metagenes for housekeeping or continually expressed genes. The remaining Inhibitors,Modulators,Libraries five metagenes corresponded to the empty vector PR Inhibitors,Modulators,Libraries null samples, and the pairwise combi nation of WT or KR PR, with or without R5020.

Thus, these analyses identified metagenes from biologically relevant subtypes of cells. Inhibitors,Modulators,Libraries The Loi et al. Inhibitors,Modulators,Libraries human breast tumor dataset contains gene expression data for both tamoxifen treated and untreated samples across several datasets. These data were aggregated together and are available through the gene expression omnibus. The dataset was loaded into Red R for proces sing. The basis matrix for the metagene analysis was reshaped to aggregate across the gene symbols and average the metagene values across each probe of the gene. The same manipulation was performed on the expression data. Non matching genes were removed from analysis. The reshaped data were supplied to the nonnegative matrix factorization package function these for scoring. As the Loi et al. data are sup plied as z scores, the data were un logged and used in the fcnnls algorithm. Samples were taken to express a meta gene if they showed a non zero value in the fitted coeffi cient matrix. Identification of novel PR target genes and comparison analysis of gene expression platforms Ligand dependent and independent PR target gene lists from two previously published studies were com bined.