EdgeR exclude 0 count subjects for each gene

Question

[RNAseqer]

Is there a way to get edgeR to take a matrix of count data containing say 10 subjects and 10 genes then perform regression for each gene ONLY using subjects with non-zero counts for that gene? So some genes may have 10 subjects, some 9 some 8 etc. This is for the purpose of investigating how 0 counts are impacting my results.

Barring that does anyone know how I would program a simple R loop to use lm() to do the same thing? Im merely intereted in comparing the genes' t-statistics with 0s in, and with 0s out.

Answer 1

[Gordon Smyth]

No, you cannot subset to different columns for different genes using edgeR.

However you can do that with limma. Just convert the counts to logCPMs, then set the values for zero counts to NA.