Scatter plots showing the two most predictive clusters of correlated probes in three independent breast cancer gene expression data sets. Each graph is a scatter plot of the negative log of the p-value for a univariate HR versus the correlation of each probe to the first PC variable derived from the expression values of the top 10 ranked probes. For each data set, only probes which had a univariate HR p-value of less than 0.05 for 5 year metastatic recurrence latencies were examined. A. Scatter plots for the 3,311 probes in the NKI2 dataset based on (left panel) unadjusted and (right panel) data adjusted for the first PC variable. B. Scatter plots for the 1,282 probes in the combined KJX64 and KJ125 datasets based on unadjusted (left panel) and data adjusted (right panel) for the first PC variable. C. Scatter plots for the 4,088 probes in the Wang dataset based on unadjusted (left panel) and data adjusted (right panel) for the first PC variable.