Fig. 1

a Simulations: From single-class data, samples are split randomly into two classes D and D*. Class effects are randomly inserted into proteins in D*. This is followed by insertion of batch effects. b PC manipulation: Proteins are sorted based on variance followed by a cutoff. The retained expression data is analyzed via principal component analysis (PCA). The first principal component (PC1) corresponds strongly to batch effects and may be removed. c Manipulation of PCs for clustering and class prediction. When combined with data of unknown labels, class labels can be predicted based on co-clustering with samples of known labels