Figure 4
Comparison of model performances based on data integrated by poe transformation (A and C) and global standardization (B and D). A set of top 10 to 200 genes were used in a classifier to construct risk index and 40thto 70thpercentiles of the cross-validated RIs were then used to dichotomize samples into a high risk or a low risk group. A. Misclassification rates based on poe transformation and B. based on global standardization. C. Performance of the 90-gene signature built on poe and D. built on global standardized data in differentiating patients at low risk of recurrence from those at high risk of recurrence.