Hi,

I am experimenting with the use of size factors in DESeq2 (i.e. the s_ij factors). Essentially, I would like to show that the use of gene-specific s_ij factors, derived from external data, improves the model's fit.

Is there a simple way to derive an AIC-like measure from my DESeq2 object for this purpose? Or perhaps there is a better method to do this?

Apologies if this has been answered before - I couldn't find anything appropriate.

Thanks,

Jonathan

You could look at the likelihood with different size factor vectors. You can put in arbitrary size factors with sizeFactors(dds) <- ... before running DESeq(). In the mcols(dds) after running DESeq() you can find the deviance which is -2 * log likelihood.

You could use cross-validation: fit the model on a subset of the samples,  then apply to the held-out samples and show that prediction performance on these increases.