Figure 4.

Predicted clinical outcome of the AML cases cited by Golub et al. (1999) as ‘unknowns’. Most cases cluster (average linkage) within the ‘therapy outcome = success’ sample group. The method opts to not make a prediction for two samples because they fall between the two classes and are not nested within either cluster. This important feature of the method will allow the identifications of cases that are outside of the classification problem, such as those that belong to a third, unidentified class.