Figure 6

Visualization of 3D embeddings for gene-expression data (breast cancer). 3D visualization of embedding results for lung cancer gene-expression data: (a) X _SSGE , (b) ${\tilde{\mathbf{X}}}_{GE}^S$, (c) ${\tilde{\mathbf{X}}}_{GE}^{US}$, (d) ${\tilde{\mathbf{X}}}_{PCA}^S$, (e) ${\tilde{\mathbf{X}}}_{PCA}^{US}$. The 3 axes correspond to the primary 3 eigenvalues obtained via different DR methods (SSAGE, consensus-GE and consensus-PCA), while the colors of the objects (red and blue) are based on known class information (cancer and non-cancer, respectively). Note the relatively poor performance of (a) semi-supervised DR compared to (b)-(e) consensus DR. Both supervised ((b) and (d)) and unsupervised ((c) and (e)) consensus DR show relatively consistent separation between the classes with distinct, tight clusters. The best clustering accuracy for this dataset was achieved by (c) unsupervised consensus GE $\left({\tilde{\mathbf{X}}}_{GE}^{US}\right)$.