Figure 7

Comparison of sample representations with regularized dual CCA and sparse CCA. Representation of the samples from the validation set by their coordinates in the first two pairs of features extracted from the tuning set using regularized dual CCA (left panel, τ _x = 0.75, τ _y = 0.3) and the sparse CCA proposed by Witten et al. [21] (right panel). The filled markers represent the coordinates in the features extracted from the copy number variables, and the open markers represent coordinates in the features extracted from the gene expression variables. Samples with different leukemia subtypes are shown with different colors. For computational reasons, the comparison is performed using a subset of the copy number and gene expression variables. The features extracted using regularized dual CCA have a higher correlation, as shown by the shorter distance between the points corresponding to each sample in the left panel. Moreover, the E2A/PBX1 group is more easily discernible using regularized dual CCA.