Figure 5

Sample classification and prediction results. (A) Leave-one-out error rates of classifiers. The KNN algorithm (blue line) reaches an optimal prediction efficiency of 95% with a minimum of 50 genes. Using 125 genes the SVM algorithm (green line) obtains this efficiency, and converges with KNN. (B) Hierarchically sample clustered (Pearson correlation metric with average linkage) profiles for the 50 gene predictor set. Notice, that although only 19 out of 20 samples were correctly classified, hierarchical clustering separates all samples into two general treatment-related clusters. (C) A summary of KNN sample classification results, showing details of the misclassification of individual samples. Although most samples classes were correctly predicted, PBMC69, an SH sample, was consistently misclassified. P = PBMC.