The median k-mer count of a read is smaller than the actual sequencing depth of the region this read represents. We have seen this before, like in the diginorm paper and the analysis of IGS method. I am not sure if we have discussed this, actually there is a way to convert such median k-mer count/ read coverage into real sequencing depth(coverage on nucleotide level).

It is mentioned in the Quake paper.
"Note that the expected coverage of a k-mer in the genome using reads of length L will be (L-k 1)/L times the expected coverage of a single nucleotide because the full k-mer must be covered by the read. "

http://genomebiology.com/2010/11/11/R116/figure/F3

I guess this has some influence to our method.

For the IGS method, now it is easier to explain the concept. The size of a IGS is exactly the length of read now, just the abundance of IGS needs to be adjusted into "real" coverage, rather than coverage from median k-mer abundance.

For diginorm, when we use normalize-by-median.py with parameter "-C" to set the "desired coverage", this actually means the coverage from median k-mer abundance, which is not the coverage based on nucleotide level.

Practically I don't think this is a big deal. But it may be worthy of some discussions.
A potential feature to consider is that we may use the "real" coverage as an optional argument. The conversion is straightforward, as shown in the formula above.