hello,
I like to quantify expression changes at enhancers of a specific TF and later correlate them to gene expression changes (same data set).
I have the positions of the enhancers from ChIP-Seq experiments and counted tags from total RNA-Seq that cover these positions.
I have a problem with normalizing these reads as they seem to change at 80% of the sites when compare to a control sample. So I can't use DESEQ. Do I now normalize by libary size, total counts at the enhancers or use DESeq scaling factor from the gene expression analysis from the same samples?
Contact Karen Wu at Lucerna Technologies, they detect transcript variants using RNA fluoro technology
Dear Karan,
thank you for your suggestion, but I am interested in a global view on enhancer transcription. I guess a fluorophor approach is more interesting when studying specific transcripts.
Dear Hassan,
thanks for the link. I got some ideas on how to approach eRNA function.
Sadly, they did not see eRNAs in there stand-specific RNA-Seq.
" However, we failed to find evidence of transcription of DHRS4L2-AS and DHRS4L1-AS, and this absence is also reflected in the strand-specific RNA-seq database of the UCSC ENCODE project."
But this might be a depth problem. I see some transcription from enhancers in my data.
regards
Franziska
Dear Franziska,
I am no specialist of eRNAs, but as you, I would have favored a solution based on raw count analysis such as EdgeR / DESeq / DESeq2. What tells you that you can't use DESeq(2)? Too much variation compared to control? Is DESeq(2) supposed to work only with a small fraction (
Dear Fabrice,
Thanks for your reply. My concerns with DESeq is the required assumption that most of the genes/eRNAs should not change there expression for propper sample normalization/model builing. For my eRNAs,I specifically look at enhancers bound by a activated TF and almost all enhancers change expression whem compared to control.
If you have total RNA-seq, can't you analyze both mRNA and eRNA at the same time, using DESeq2 (assuming the majority of the mRNA doesn't vary)? I have no idea of the relative numbers of reads / TSS between mRNA and eRNA, but if you have enough expression-stable mRNA in your data, they might counterbalance the 80% of varying eRNAs? Then you'll subtract information about eRNAs only, but at least the statistical paradigm of DESeq2 will be fulfilled. It's a bit of a wild guess...
Good idea! I used Deseq for the analyis of differentially expressed mRNAs from the same data. Then got the scaling factors to normalize the eRNA counts for differences in libaries.Looks good so far.
Thank you.
Hi Franziska, you are correct about the fact that normalization in DESeq might be not be most suitable, if all the eRNAs undergo a global change. I would try two approach
1. is to normalize by library size. For example if I have 3 libraries with total read counts N1, N2 and N3. A quick normalization factor would be N1/mean(N1,N2,N3) , N2/mean(N1,N2,N3).. In R it would be sz = mean(c(N1,N2,N3).
So in DESeq2, you can specify the size factors like sizeFactors(dds) = sz
2. To treat each eRNA like a gene and add them to your gene count table. Then run DESeq on this combined gene table. This would be similar to using the scaling factor to normalize except you might have a more accurate estimate of the variation of eRNA..
Either way, very interesting question you have and hope the suggestions help!
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Biology
- Biomolecules
- Nucleic Acids
- RNA