Sequencing depth taken into account with FPKM?

[aplany]

I had a question regarding FPKM calculations.

I've sequenced transcriptomes for two different species and am comparing their response to a specific treatment. Both species have very similar genome sizes. However, I sequenced about 40 million 100bp-PE reads for one of the species and ~33 million for the other.

I've done the mapping against their respective genomes using bowtie. However, I'm now concerned about coverage differences.

When I am doing transcript quantification using RSEM, do I need to first randomly subsample the reads of the first species so I have the same depth of coverage going into quantititation, or will FPKM calculations take the coverage into account?

[HESmith]

Do you know what FPKM stands for? There's your answer.

[rflrob]

Less acerbicly, FPKM is fragments per kilobase per million mapped reads. That million mapped reads in the denominator should do an okay job of dealing with the coverage differences. The idea is that, for a given organism, you can go from FPKM to copy number per cell, only making a handful of overly-broad, almost-believable assumptions.

It's worth bearing in mind, though, that simply comparing FPKMs between two samples won't do a very good job of pulling out statistically significant differences. For that, you should look into a program like DESeq, Cuffdiff, or edgeR.