Hello everyone,

I have 18 soil metagenomes sampled from three different conditions (6 samples each), for which I have assembled. I’d like to do a comparative study of the genes that could potentially be expressed in these samples. What I did was that I first estimated the presence of KOs using KEGG’s kofam database on the contigs, and I also got the coverage of each KO in my samples. Next, I calculated the total coverage of each KO (i.e. by summing the coverage of the same KOs on different contigs) in every individual sample. I am wondering if I can use these coverage values as input for DESeq2 as opposed to read counts mapped on each gene?

An example of my input data would be

..... the list continues

Thank you for any insights!

Rui

It's not ideal. Basically, DESeq2 is attempting to model the data given that it consists of observations of reads which have a certain variance property. Normalized counts, or coverage data, do not have that property exactly and so the method won't perform as well as it would with the proper input. I've seen this when people provide normalized counts to DESeq2 instead of original counts -- the method does not have the ability to associate higher variance on the estimated LFC from low counts, when those low counts have actually been scaled up to an artificially high value.