Figure 3

Scaled principal components for samples in toxicology data measured by Spicker et al. (2008), found in various training data sets [15]. Groups 1, 3, and 5 were those for carcinogens, groups 2, 4, 6 were those for nontoxic compounds, and C was the mock control. A. Training data was prepared by using all of the groups' sample means. B. To mimic bias, the sample means of group 3 were replaced with all the samples. The increased sensitivity towards the group 3 samples is obvious, especially in the second axis. C. Training data were prepared without group 3, and the data of group 3 were subjected as unknown samples. The carcinogens including group 3 were separated from the nontoxic groups. D. An example of biplot-like presentation. Scaled principal components of the training data for genes and groups are presented in a same scale.