Hello,

I am analyzing TCGA data, RNA SEQ, of tumoral tissues. I would like to perform a correlation analysis with gene expression (to see if gene expression of Gene 1 correlates with Gene 2 in the sample, for example) and some clinical data (like alpha-fetoprotein levels, age, bilirubin levels...).

My doubt is: should I use FPKM data or normalized counts generated by Deseq2? Or something else?

Thank you!

If you want to use Spearman correlation, the SAMseq function implements this with resp.type="Quantitative"

https://www.rdocumentation.org/packages/samr/versions/3.0/topics/SAMseq

In DESeq2 you could add numeric covariates to the design, which assumes that unit changes in the covariate correspond to constant fold changes in the counts.

You may also use the edgeR package, whose starting point are also raw integer counts. Once you've built a DGEList object and calculated normalization factors with calcNormFactors(), the function cpm() can provide you continuous log-CPM units of expression suitable to be used for clustering and other gene-correlation purposes (see subsection 2.16 from the edgeR User's Guide). As a side note, you might also want to look at this preprint, which investigates proper ways of calculating correlations between genes in RNA-seq data, providing an R package called spqn that implements the approach in the preprint. Note, however, that the package is still not in Bioconductor and the preprint has still not gone through peer-review and therefore, you would have to contact directly the authors to get support in using their method, e.g., opening an issue in the GitHub repo.