PCA analysis of a MALDI-imaging dataset. Illustration of shortcomings of PCA analysis of a MALDI-imaging dataset of a rat brain coronal section of 10 μm thickness (160 μm spatial resolution, 101×60 pixels, 5053 spectra). A. Score images of first six principal components (PC). B. For each score image, the m/z-image with the highest loading is plotted. One can see that the visual correlation between a score image (left) and its highest-loading m/z-image (right) is achieved only for PCs 1, 2, and 5, whereas not achieved for PCs 3, 4, and 6. Thus, the use of loadings for interpreting score images is not recommended.
Alexandrov BMC Bioinformatics 2012 13(Suppl 16):S11 doi:10.1186/1471-2105-13-S16-S11