Open laopp opened 6 years ago
Under the best of conditions RNA-based copy number inference is only accurate for large-scale alterations, and these are not the best of conditions. But if you were able to run RSEM to quantify this RNA-seq data then it should work to some extent with import-rna
. Do the RNA and DNA-based calls at least agree at the chromosome arm level? Or are the RNA results pure noise?
Here are the heatmaps from DNA seq data and RNA data.
The samples are in the same order except that normal is included in the RNA results at the very bottom. You can see that there are some degree of consistency for 1q and 8, but wonder why RNA data is so blue in general. Also we would expect normal to be close to zero, but it turns out to be very negative. Any thoughts?
It looks like import-rna
didn't filter out low-expressed genes well enough. That's easy enough to fix with the call
command:
cnvkit.py call DCIS5_1.cns -m none --center -o DCIS5_1.centered.cns
Once you do that for each RNA sample, I think at least the gain at 8p should show up in DCIS5 and 6.
It wouldn't hurt to do the same recentering for the DNA samples, too. I see a lot of light blue in the DNA heatmap along with large-scale gains that are probably real and would show up more clearly if all log2 ratios were shifted up a bit.
Yes the centering on RNA samples is very helpful. The gain at 1q and 8p does show up on the heatmap. But I wonder why normal sample does not show 0 across all chromosomes, are the samples normalized against the normal samples?
I also did centering on the DNA samples, and I don't see much improvement. some light blue gets even lighter after the centering.
Currently the -n
option in import-rna
doesn't do anything -- it's an unimplemented feature that I probably should have hidden before pushing out the last release. (Or finished implementing it.)
If the majority of your RNA samples have the same copy number alteration, then that alteration will be treated as the baseline and your normal/control sample will show a spurious alteration in the opposite direction. Is that what happened here?
The -n/--normal option is in place now (#362), if you'd like to try it out.
Hi Eric,
Thanks for letting me know. Will definitely give it a try!
Thanks,
Peipei
On Thu, Jun 7, 2018 at 4:46 PM Eric Talevich notifications@github.com wrote:
The -n/--normal option is in place now (#362 https://github.com/etal/cnvkit/issues/362), if you'd like to try it out.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/etal/cnvkit/issues/352#issuecomment-395600577, or mute the thread https://github.com/notifications/unsubscribe-auth/AlWoQpvVEnuyGh66nAkzx36ayDQCG6qKks5t6btygaJpZM4T6TH4 .
Hi Eric, I tried the --normal option in import-rna. It seems that the blue/red are getting even more lighter after the normalization. Also, should I expect that the normal sample has a near 0 baseline on the heatmap? Obviously I didn't see that. Thanks!
Hi Eric,
Thanks for implementing this. I'm also wondering about the issue above. I wouldn't expect to see such gains or losses in the normal sample(s). Any thoughts on what could be causing this result?
Sujay
Not really sure, this part of the code is less tested and could use another set of eyes on it.
Hi I'm trying to run import-rna with raw read counts. The results are very different from that by targeted DNA seq. The samples for RNA seq vs. DNA seq are 99% same so they are supposed to generate similar results. My question is that our samples are FFPE (formalin-fixed) archival tissues and sequenced by the 3'-end sequencing methods rather than the traditional RNA seq. The method generates a directional sequencing library comprised primarily of ~200 bp cDNA fragments with a poly-A tail, in which sequencing will proceed directionally toward the poly-A tail. Do you think that import-rna will work on the 3’-seq data? Thanks!