Relationship coverage/k-mer

[Antony03]

I tried to test an assembly with different k-mer (11 to 31 because I have very shorts reads) with the de novo assembler Ray. My dataset is from a sequencing project with Illumina technology. There is something who looks strange to me.
I made a graph of the coverage per contigs (50 biggest but without the first because their coverage are to high and make hard to visualise clearly the graph).

My results suggest that more high is the k-mer more the coverage is important and looks like a gaussian. Longer k-mers are inherently rarer, so my results don't make sens...
Is there a person with a explaination?

This is a screenshot of graph:

http://dl.dropbox.com/u/61694030/illumina.png

Thanks!

[luc]

hi,
this might be the wrong screenshot - I see a mac laptop.

[Antony03]

Sorry! A problem with my dropbox. I changed the url in the last post and this is the new url for the screenshot!

http://dl.dropbox.com/u/61694030/illumina.png

Sorry again.

[Antony03]

Up please