Table 4. Class prediction using Prediction Analysis of Microarrays: Consistency between hierarchical clustering and prediction for the three data sets.				
No. of tumor classes*	Training set 	Test set	No. of genesà	Accuracy¤
5	Norway/Stanford	vanÕt Veer	97	0.79
		West	58	0.88
4	Norway/Stanford	vanÕt Veer	101	0.89
		West	54	0.88
"*A five-class predictor (luminal A, luminal B, basal, ERBB2, normal-like) or a four-class predictor (luminal, basal, ERBB2, normal-like) was trained on the Norway/Stanford data."				
 The Norway/Stanford data set of 115 malignant tumors (122 experiments) were used to train the classifier.				
àDifferent subsets of genes were optimal for the two different predictors in the two different data sets.				
¤Correlation between hierarchical clustering and statistical prediction.