Figure 4

prostate cancer prediction. Sensitivities (a) and specificities (b) (with standard deviations as error bars) estimated in prostate cancer prediction from protein expression levels using 100 independent two-fold cross-validations and linear SVM classifier. Four sets of selected components were extracted by SCA-based factorization using LMMs (2a) and (2b) with control reference (c.r.) and disease reference (d.r.) samples respectively, where the overall number of components M has been set to 2 (red bars), 3 (green bars), 4 (blue bars) and 5 (magenta bars). Optimal values of the parameters λ and Δθwere used for each M. Performance improvement is visible when number of components is increased from 2 to 5.