Fig. 2

Effect of feature specific normalization methods on test and training breast distributions. Left block (Dark Blue): Normalization using RNAseq data as training distribution. Right block (Gold): Normalization using microarray data as training distribution. Colour legends for each block are provided. a, g. Probability density functions of log₂ microarray and RNAseq data prior to feature specific normalization. b, h Principal Component Analysis (PCA) plots of log₂ microarray data and log₂ RNAseq data. The first (PC1) and second (PC2) principal components are projected on the x-axis and y-axis, respectively. c, i PCA plot of the first two principal components of gene expression data after feature specific quantile normalization (orange) to the respective training distribution (blue) demonstrates limited variation between gene expression platforms after FSQN. d, j. PCA plot of the first two principal components of gene expression data after feature specific mean–variance normalization (green) to training distribution (blue) demonstrates limited variation between gene expression platforms after FSMVN. e, k. The probability density function of gene expression data after FSQN demonstrates the shift of the test distribution (orange) to match the training distribution (blue). f, l. The probability density function of gene expression data after FSMVN demonstrates the shift of the test distribution (green) to match the training distribution (blue)