EdgeR Fold Change Calculation

[umnklang]

Could anyone help me understand how edgeR calculates fold change?

Experimental set up:
RNA-Seq using illumina with 2 identical bacterial strains - one wild-type and one with a single gene deletion. We did not sequence biological replicates for this experiment so I only have one set of count data for each strain.

I set the dispersion to 0.1 as outlined in the EdgeR materials for identical strains.

I get the following output for a gene and I don't understand why

GeneID WTcount mutantcount logFC logCPM p-value
7872684 2291 7671 -0.271135002 13.11904664 0.679110027

This should be ~3-fold increase, definitely not a decrease. Maybe I'm missing something.

[chadn737]

1) You need biological replicates, unless this is simply a pilot experiment of sorts, you cannot rely on single samples to accurately detect differential expression.
2) Look at your p-value, you can't say that gene is differentially expressed
3) How many reads per sample do you have? You can't simply compare raw read counts and say that this is the difference. First you have to normalize the data and account for things like differences in numbers of reads. If your Mutant has a disproportionately larger number of reads, then of course it will have more reads than the WT before normalization, even if there is actually no difference.

[umnklang]

Thank you for your reply; I've just realized I have nearly 10x the number of reads that map to my CDS's for my mutant strain. This explains why I'm getting this result. Makes a lot of sense now.

mahalo

[colindaven]

There is a library normalisation step in edgeR which is relevant for this problem. Have a look at their very nice PDF tutorials!