I have been studying the package WGCNA in R. This package was developed with the idea that one could cluster co-expression values of genes from microarray data to form modules within the gene-networks... While I understand the principle of PCA and eigenvalues/ vectors, the papers that describe the methodology as though it extracts the major "axes" of variability.... How can there be more than one axis of variability when they are just expression values? I realise that an "eigengene" is the summarisation of the expression levels of a module- but doesn't that just mean that it is the average/spread of expression of each module? So I'm wondering how can one reduce dimensions in such a dataset when there is only one dimension to consider- expression values?
You have as many dimensions as you do genes/transcripts/whatever.
Thanks Devon, so does that mean that in reducing dimensions one is revealing the informative genes?
I wouldn't say that one is revealing the informative genes, rather one is reducing the size of the dataset to a more manageable size.