A. Procedure for detection of informative genes and class prediction. The ranked expression profile of gene g (E'g) corresponds to a labeling vector (V'g) which contains two regions that group together a significant number of class 1 and class 0 samples (50% and 75%). These regions map on two CERs: [39-50] and [58-86], respectively. The gene can thus be used for the classification of a new sample S1 in which its expression lies between the respective thresholds of CER2, but not for a sample S2. B. Examples of labeling vectors -that correspond to informative genes- identified by our method. Vectors V and VI are of particular biological interest as they divide samples of class 0 and 1 respectively, in two distinct sub-groups.
Pavlidis and Poirazi BMC Bioinformatics 2006 7:345 doi:10.1186/1471-2105-7-345