I am interested in the possibility of using PCA or other such segmentation approaches to examine whether functionally related components of large multiprotein complexes or signaling pathways are coregulated in cancers. So for example are there specific sub-complexes of the nuclear pore complex or ribosomes that are reduced or increased in amounts in particular types of cancer? The question was inspired by observations that the ribosome, for example, has components that are used in different tissue types. I have downloaded ~9000 cancer transcript samples from the TCGA, normalized the reads, and then carried out PCA on the standardized data. There is, perhaps not surprisingly, little evidence of any structure in the results. As the next step, it might be better to go tumor type by tumor type and compare them with a suitable control. Given problems with batch effects and the apparent lack of normal controls on the TCGA database; could somebody suggest a suitable source of STAR protocol controls? At a first go, they do not even have to be tissue matched; I would just like to know; does one see any structure to the output? Is this approach practical?
