DML Analysis Part 46
Epigenetics reading group feedback!
Although the whole gang wasn’t present Mac and Steven gave me some great feedback on the C. virginica edits I started on! Here are some of the highlights:
- Clearly defining what I consider a DML, since a locus doesn’t have to be one base pair
- Rephrasing GO-MWU input creation methods
- Not using signed p-values for GO-MWU. Mac made a good point that understanding how up- or downregulated genes influence enrichment patterns makes more sense for transcription than seeing how hyper- and hypomethylated DML influence enrichment. We don’t know what methylation is doing in a gene, and using signed p-values doesn’t shed any light on that function. Additionally, hypermethylation could lead to increased or decreased gene activity. If that gene is responsible for repressing another gene function, assigning a positive or negative value based solely on methylation status doesn’t take into account gene function.
- Removing DMG analysis since the actual function of methylation in these genes is unknown (no paired transcription data) and it didn’t add anything to the paper.
- I moved the percent methylation across genome feature figure to Figure 1 instead of keeping it with Figure 2.
- Changed Figure 8 so it was colored by gene group, not biological process
Calculating “genome methylation”
In the paper, I state that 22% of the C. virginica genome was methylated. Mac and Steven were hesitant to make this claim since it wasn’t clear how I got this number, and I couldn’t find it in the methods anywhere. When I looked at the 3,181,904 methylated CpGs (> 50% methylated) versus the 14,458,703 total CpGs in the C. virginca gneome, I got 22%. This doesn’t saying that 22% of the genome is methylated. Instead, 22% of CpGs were found to be methylated. In this Jupyter notebook I counted the number of cytosines in the C. virginica genome. Out of the 119,229,194 cytosines in the genome, 3,181,904 methylated CpGs means that the 2.7% of cytosines are methylated. This is similar to what Mac reported in her 2013 paper. I clarified this statement in the abstract, results, and discussion.
I removed the methylation island bar in Figure 2, since the methylated CpGs and methylation islands are correlated. Mac described methylation islands as a smoothing function, so it’s not really interesting to see where they are located with respect to the other genome features beyond just knowing what’s genic and intergenic. I also removed Figure 2b (individual feature locations in methylation islands). After calculating median and ranges for methylation island length and number of mCpGs, I tried creating a histogram of island lengths in this R Markdown file. When I created a preliminary plot, it was clear that I would need a gapped axis! I previously created a barplot with a gapped axis, so I thought I could recycle code. I kept getting the following error when I tried plotting my data:
Error in rect(xtics[littleones] - halfwidth, botgap, xtics[littleones] + : cannot mix zero-length and non-zero-length coordinates
In the meantime, I looked at the histogram bins and how many islands were in each bin. I wrote up a quick description for the results section.
Figure 1 and Figure 5 needed a little extra love when converting them into robust multipanel plots. In this R Markdown file, I combined the CpG frequency distribution I made previously with the plots of median percent methylation globally and for each sample across genome features. It took some trial-and-error to get the spacing to look the way I wanted it to, but I like the final product!
Mac suggested I convert the multiple barplots in Figure 5 (panels c-f) to stacked barplots. Turns out this would be the bane of my existence. I spent a bit doing some standard dataframe manipulation to count the number of genes with DML that had 0, 1, 2, 3, or 4 hypermethylated DML. I figured if I just considered data across different DML totals by the number of hypermethylated DML, I could always derive the number of hypomethylated DML by subtracting the number of hypermethylated DML. I performed these manipulations for genes that had 2, 3, 4, and 5 DML, creating a stacked barplot that replaced the four original panels.
When I went to edit the figure caption in the manuscript, trying to describe this layout and figure was difficult. That made me realize I needed to rework the figure one more time. I decided to go for a fully vertical layout with chromosome and gene information on top, number of DML per gene in the middle, and breakdown of hyper/hypomethylated DML on the bottom. The panel with number of DML per gene had information for genes with 1-5 DML. I decided to mirror that x-axis in the bottom plot with the DML type.
Writing the caption for this figure was much easier, so I think this is the way to go.
Things to improve in the discussion
- Look at list of genes with DML and see if any of them have been reported in ocean acidification and gene expression studies (the biomineralization and protein ubiquitination stuff should definitely have associated gene expression studies, so this may be a good place to start)
- Add ocean acidification context to final two discussion paragraphs, along with caveat about experimental design confounding responses to ocean acidification with potential differences in gamete maturation
- Address remaining comments about discussion text
- Update manuscript text
- Update response to reviewers
- Consolidate any co-author feedback
- Submit comment responses and reviesed manuscript
- Post revised paper on bioRXiv
- Update paper repository with new code and figures