I was reading this paper related to use of gene expressions for predicting the drug response. I have this confusion, the paper has used PCA on the covariance matrix formed by the genes to get what is called the background network as they say.

I don't know the essence of background in this case. I know PCA gives us components which preserves the greatest variance and used for dimension reduction. What they mean by background network. Is it pathways ?

  • 1
    $\begingroup$ For reference, previous question regarding this paper: biology.stackexchange.com/questions/7800/… $\endgroup$
    – blep
    Commented Apr 16, 2013 at 22:48
  • 2
    $\begingroup$ Could you include a bit more information on the paper so we can understand more easily? $\endgroup$
    – terdon
    Commented Apr 17, 2013 at 18:43
  • $\begingroup$ Perhaps, the background is the set of genes which are not represented by the first few principal components $\endgroup$
    Commented Apr 18, 2013 at 5:02

1 Answer 1


I am by far not an expert, but I read the paper and maybe I can help clarifing things a bit.

Let's start with the simplest answer to your question: In the last paragraph of the Introduction paper the authors say

"Throughout this manuscript [...]'background networks' and principle components are used interchangeably." ([1] Torkamani and Schork, Genetic background and drug response, The Pharmacogenomics Journal (2012) 12, 446-452; doi:10.1038/tpj.2011.35)

So they did not "use" PCA to get the background networks, they just call the PCs background models to indicate how they interpret it: As stated earlier in the introduction

"...interacting networks cannot be expected to correlate strongly with drug response, as their influence may only be observed when the major determinant of drug response and the interacting network complement one another or are both at a synergistic state. A major problem with identifying these interaction partners [...] is the extremely large number of possible partners [...] and [...] that individual genes are unlikely to accurately represent the overall state of a biological network." [1]

As far as I understand the article, they suggest PCA as a kind of compromise: Ignoring interactions would miss associations of all networks lacking a single gene representing the network's state accurate enough. Including all interactions is infeasable due to the huge number of gene-pairs. By PCA, the number of interactions can be decreased by orders of magnitude while keeping a maximum on information (as in variation): Instead of using the interaction (as in product) of all probes with all probes, they only consider those of all probes with the first six PCs.

I think, in this context 'background' is not used in the meaning of 'background noise' but of 'cultural background' - instead of assesing individual interactions, genes are assigned a 'background'. Since this corresponds to a) the biological network(s) they belong to and b) the principle components(s) they contribute to, those two ideas are used synonymously by calling PCs 'background networks'.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .