Figure 4.

A stochastic universal cancer classifier. (A) ROC curves for a 10-fold cross-validation experiment classifying any sample as normal or tumor, where the anti-profile is trained (genes selected and normal regions of expression defined) independently for each fold, and the ROC is computed for each testing fold independently. (B) ROC curves for 7 leave-one-tissue-out experiments. In each of the leave-one-tissue-out experiments, all samples of that particular type (both normal and tumor) are removed from training sets and then scored using the resulting anti-profiles. (C) Cross-validated anti-profile scores for the 7 leave-one-tissue-out experiments. The anti-profile scores can separate a large number of tumors from their corresponding normal samples.

Corrada Bravo et al. BMC Bioinformatics 2012 13:272   doi:10.1186/1471-2105-13-272
Download authors' original image