Hawaii Gigas Methylation Analysis Part 6

Preliminary methylation investigation

While my files are aligning to the new genome, I want to use the files I have to test out some workflows! First on the docket is getting a preliminary methylation assessment from methylKit.

Fixing genome preparation parameters

But first…my bismark script failed RIP. Looking at the slurm file, I saw this error:

Screen Shot 2021-03-04 at 8 43 52 AM

I used --path_to_bowtie in previous genome preparations and it’s an argument for the alignment step, so I was a little confused. I looked at the Bismark User Guide, I saw that --path_to_aligner is the argument that should be used instead! I made that change in this script, then started running it again.

Then I ran into my second problem:

Screen Shot 2021-03-04 at 10 44 28 AM

NC_001276.1 is the mitochondrial sequence ID, and it’s at the end of the genome. Since bismark is seeing repeated chromosome numbers, I thought it must be pulling information from the other two FASTA files in that folder: the original Roslin genome and the separate mitochondrial sequence. I removed these files from the directory, and used rsync to move the combined genome file I created to this location on gannet. I reran my script, and it started running with no errors! I checked on it after some time, and found that the genome was fully prepared and that it was starting to align samples to the bisulfite converted genome.

Methylation assessment

Now that I knew bismark was running, I wanted to complete a preliminary methylation assessment. This would not only allow me to play with methylKit code (it’s been a while), but also get an understanding for any differences in methylation between diploid and triploid oysters, or between treatments. I definitely want to use the PCA methylKit produces in the manuscript, and look at the methylation histograms and correlation plot it produces! There’s definitely a better way to identify DML than two pairwise comparisons, but if there aren’t any differences between ploidy status or OA treatment, then I can treat the experiment like a single-variable study.

To start, I created this R Markdown script. I downloaded the merged CpG coverage files from gannet to use in the analysis, similar to what Mac did for the MethCompare project. After downloading the files, I used methRead to import the coverage files. I previously used processBismarkAlign, but I’m working with merged coverage files this time, so I used methRead instead. I still used filterByCoverage to ensure that there was a minimum 5x coverage for each locus, and I excluded data with coverage in the 99.9th percentile:

setwd(dir = "/Users/yaamini/Documents/project-oyster-oa/analyses/Haws_04-methylKit/") #Set working directory inside code chunk processedFiles <- methylKit::methRead(analysisFiles, sample.id = list("2H-1", "2H-2", "2H-3", "2H-4", "2H-5", "2H-6", "2L-1", "2L-2", "2L-3", "2L-4", "2L-5", "2L-6", "3H-1", "3H-2", "3H-3", "3H-4", "3H-5", "3H-6", "3L-1", "3L-2", "3L-3", "3L-4", "3L-5", "3L-6"), assembly = "oyster_v9", treatment = c(rep(0, times = 12), rep(1, times = 12)), pipeline = "bismarkCoverage", mincov = 2) #Process files. Treatment specified based on ploidy status. Use mincov = 2 to quickly process reads. 
processedFilteredFilesCov5 <- methylKit::filterByCoverage(processedFiles, lo.count = 5, lo.perc = NULL, high.count = NULL, high.perc = 99.9) #Filter coverage information for minimum 5x coverage, and remove PCR duplicates by excluding data in the 99.9th percentile of coverage with hi.perc = 99.9. 

After each of these steps, I used save.image to save the R data. R Studio crashed very frequently when I was testing out code, and I wanted to save the output of these time-consuming steps!

I looked at descriptive and comparative statistics, and saved any plots in this folder. For the descriptive statistics, I got sample-specific CpG coverage and methylation information. Skimming through the plots, I saw the standard spikes in CpG methylation at 0%, and a super smol spike the end of the distribution. Coverage looked relatively consistent, but sample 7 had more reads covered at higher depths than other samples. I produced a correlation plot, clustering diagram, PCA, and scree plot for the comparative analysis. The clustering diagram didn’t reveal any ploidy or treatment similarities, and neither did the PCA. Sample concordance was at least 85%, which further supports the idea that there aren’t ve3ry many differences in methylation patterns between samples. I’m curious to see how removing SNP data will affect the analysis. I also saw two potential outliers in the PCA (2H-1, 3H-2), so I’ll also need to make sure that the DML have data represented in at least 10/12 samples (or something like that).

Going forward

  1. Create covariate matrix and complete pairwise DML assessment in methylKit
  2. Try BS-SNPer and EpiDiverse for SNP extraction from WGBS data
  3. Investigate comparison mechanisms for samples with different ploidy in oysters and other taxa
  4. Test-run DSS and ramwas
  5. Transfer scripts used to a nextflow workflow
  6. Update methods
  7. Update results

Please enable JavaScript to view the comments powered by Disqus.

from the responsible grad student https://ift.tt/30gQ3nA
via IFTTT