Figure 4.

Predictive model for classifying tumor samples for the indicated datasets. The predictive model was evaluated for class assignment by transcriptome signature in all datasets except the Tschentscher dataset, which was evaluated for monosomy 3. Probe sets were entered randomly (−mRMR) or by minimum redundancy and maximum relevance (+mRMR) into a weighted voting algorithm. Classification errors calculated by leave-one-out cross-validation are plotted on the upper graph. The mean and minimum confidence scores are plotted on the lower graph.