Figure 4
From: Intrinsic bias in breast cancer gene expression data sets
ER positive tumors depend upon proliferation genes for cluster assignments with significant differences in latencies. A. This graph was based on unsupervised hierarchical clustering analyses using the 5,000 randomly-generated unigene gene. The graph shows the percentage of lists that were significantly predictive of metastasis recurrence latencies in ER positive tumors in either unadjusted data sets (grey bars) or data sets that were globally adjusted for proliferation (black bars). B. The graph shows the percentage decrease in the sum of the variances of all genes in ER positive tumors after globally adjusting each data set for a marker of proliferation.