Hi all,
I have analysed some RNA-Seq data using the DESeq2 protocol for analysis of time course experiments between two conditions (http://master.bioconductor.org/packages/release/workflows/vignettes/rnaseqGene/inst/doc/rnaseqGene.html#time-course-experiments).
Now I have collected samples for MNase-Seq under an identical experimental setup and want to compare how the nucleosome occupancy is affected between my 2 conditions during the time-course. I have the count of adjusted reads for summits (i.e nucleosomes) calculated using DANPOS. Is it correct to to apply the DESeq2 likelihood ratio test to the MNase-Seq data, in order to identify changes in occupancy between the two conditions during the time-course?
Can I use these values to build a counts matrix and calculate differential summit values during the time-course between the two conditions or is there other package more adequate for MNase-Seq time-course analysis?
Many thanks!
Dear Michael,
Thanks for your reply.
According to the paper where DANPOS was described https://www.ncbi.nlm.nih.gov/pubmed/23193179: "nucleosome occupancy is calculated as the counts of adjusted reads covering each base pair in the genome".
Adjusted because the read size is adjusted to half of the nucleosome size (74 nt) to enhance the signal to noise ratio. Also, before the size adjustment reads are shifted toward the 3' direction for half of the estimated fragment size (to compensate for the variable size of fragments after MNase digestion). Fig. 1A of the paper has a nice schematic illustration.
Regarding normalization, I don't think any "control" peaks are being specified.
DANPOS has a function that allows the identification of differential nucleosome occupancy between conditions using a Poisson test. I ran this analysis to compare my two conditions for each time-point and most of the peaks (nucleosomes) do not have changes in occupancy. I would say that 2-3% of the peaks have significant changes in occupancy between the two conditions (according to a FDR value of 0.05).
I hope this helps.
Miguel
I wouldn't recommend to use DESeq2 to model the adjusted counts, without knowing more about whether that's a good idea. Typically we just model original counts or estimated counts in DESeq2. Maybe you could log transform and use linear models in limma.