KamilSJaron
commented
4 years ago Hello @jafors,
thanks for trying out smudgeplots.
I wanted to try smudgeplot on some WGS data of murine cancer cells, and thought I'd start with a normal control sample taken from the mouse tail. So, the sample should definitely be diploid. It's WGS, and the number of k-mers was pretty large, so I decided to run every chromosome by itself to stop smudgeplot hetkmers from running into memory troubles.
Nice. However, I do wonder if there will enough genetic variation within the cancer lineage to show "heterozygous" loci. I suppose that the origin of cancer polyploid cells will be kind of whole-genome duplication, is that correct? Because in the case there might be a problem that they would be too homozygous to get a genome-wide signal. And it relates to why it gets a wrong estimate for the healthy tissue. I am also intrigued by your chromosome-wise computation decomposition, is it by mapping reads?
Now, the problem is, though we expect polyploidy in the cancer celllines, the control should be diploid. The smudgeplots of the various chromosomes show varying ploidies from di- to tetra- or even higher ploidies.
That sounds like a problem. We originally designed and tested smudgeplots on very heterozygous species with various high ploidy levels and for those, it worked very well. However, for multiple species with low heterozygosity, we are getting mixed signals of heterozygous loci and paralogs leading to smudgeplot as you show here. Where there are more duplicates (AABB) than heterozygous kmers (AB). Note that "estimated ploidy" is a very naive thing, it simply tells us what is the ploidy level with most of the kmer pairs. The visualisation itself is far more important than the estimate.
@tbenavi1 is now actively working on better smudgeplot interpretations using simulation data, but I don't think it's a production-ready code right now.
The L and U cutoffs that were suggested using smudgeplot cutoff are also pretty low compared to what you recommend in the Readme (L≈12, U≈420).
The cutoffs depend on the quality of your data and your data looks beautiful, therefore the L cutoff is fine. Having too low cutoff leads to a different pattern than the one you see (too many paralogs).
I attached smugeplot and genomescope plot for chr1. Do you have any suggestions? Is this kind of input data feasible for smudgeplot? Should I tweak the cutoffs around?
For your application in particular, where you know for sure the genome, I would say the best chance is to compare the healthy diploid with cancer. Without using the ploidy estimate directly. Now, that you know that ~56% of your kmer pairs are paralogs and 40% are heterozygous kmers, you can check if that changes with a cancer data. You might see already the effect on the kmer spectra. I would expect inflation between 0 - 60x and smaller homozygous peak. But perhaps I am too naive on how the cancer genomes look like.
jafors
mflevine
For the tumor, the ploidy states don't totally match up with the states from the copy number caller. I also had to adjust the bins.
Here is the profile from the copy number caller:
Hi!
First of all, thanks for your work with smudgeplot. Great idea!
I wanted to try smudgeplot on some WGS data of murine cancer cells, and thought I'd start with a normal control sample taken from the mouse tail. So, the sample should definitely be diploid. It's WGS, and the number of k-mers was pretty large, so I decided to run every chromosome by itself to stop smudgeplot hetkmers from running into memory troubles.
Now, the problem is, though we expect polyploidy in the cancer celllines, the control should be diploid. The smudgeplots of the various chromosomes show varying ploidies from di- to tetra- or even higher ploidies.
The L and U cutoffs that were suggested using smudgeplot cutoff are also pretty low compared to what you recommend in the Readme (L≈12, U≈420).
I attached smugeplot and genomescope plot for chr1. Do you have any suggestions? Is this kind of input data feasible for smudgeplot? Should I tweak the cutoffs around?
Best, Jan