Figure 1

Principal component analysis and biplot. Projection of five-dimensional patient biomarker profiles with no missing data (439 instances) and the 439-dimensional biomarkers profiles onto the two leading principal components of a matrix consisting of expression profiles of Her2/neu, estrogen receptor (ER), progesterone receptor (PR), Bcl-2 and Bcl-2 antanogene-1 (Bag-1) present in all 439 samples. Each patient is represented by a distinct symbol (•, alive at 10 years; ×, dead at 10 years). The accumulated variation captured by the first and second principal components is 92% of the total variation. Overlaying a two-dimensional scatter plot representing the projection of the biomarkers () onto the first and second principal components on top of the two-dimensional patient scatter plot representing the projection of their five-dimensional biomarker profiles onto the two leading principal components forms a biplot. The biplot can be used to read the approximated transformed expression levels.