Figure 2

The MDA and MDAhet classifier. Four two-dimensional projections of the seven-dimensional Mixture Discriminant Analysis (MDA) and Heterogeneous Mixture Discriminant Analysis (MDAhet) classifiers. Scatterplots show projections of the training expression data (183 oestrogen receptor negative samples) onto arbitrarily chosen two-dimensional subspaces spanned by the genes HLA-F and IGLC2, LY9 and TNFRSF17, SPP1 and XCL2, and IGLC2 and C1QA. Codings: black = poor outcome, grey = good outcome, triangle = training samples classified into the good prognosis subgroup defined by overexpression of seven-gene module 'good-up', circle = training samples not classified into 'good-up' group. In addition, the means and covariance-curves of the two Gaussians that approximate each of the poor (black ellipses) and good outcome (grey ellipses) classes are shown. C1QA = complement component 1, q subcomponent, A chain; HLA-F = major histocompatibility complex, class I, F; IGLC2 = immunoglobulin lambda constant 2; LY9 = lymphocyte antigen 9; TNFRSF17 = tumour necrosis factor receptor superfamily member 17; SPP1 = secreted phosphoprotein 1 (osteopontin); XCL2 = chemokine (C motif) ligand 2.