Question: How to do DMR analysis with BiSeq (for RRBS methylation data)?
1
gravatar for ahmad mousavi
20 months ago by
ahmad mousavi450
Royan Institute, Tehran, Iran
ahmad mousavi450 wrote:

Hi

I have 20 RRBS samples which I wanted to do do deferentially methylated region (DMR) analysis with BiSeq or other R packages, however the manual resource is not enough. I want to do DMR analysis on all chromosomes not specific regions.

Is there anybody who has experience with BiSeq:

> readBismark(files, colData)
> metadata <- list(Sequencer = "Instrument", Year = "2013")
> rowRanges <- GRanges(seqnames = "chr1",ranges = IRanges(start = c(1,2,3), end = c(1,2,3)))
> colData <- DataFrame(group = c("cancer", "control"), row.names = c("sample_1", "sample_2"))
> totalReads <- matrix(c(rep(10L, 3), rep(5L, 3)), ncol = 2)
> methReads <- matrix(c(rep(5L, 3), rep(5L, 3)), ncol = 2)
> BSraw(metadata = metadata,rowRanges = rowRanges,colData = colData,totalReads = totalReads,methReads = methReads)
sequencing dmr methylation • 1.0k views
ADD COMMENTlink modified 7 months ago by Charles Warden7.2k • written 20 months ago by ahmad mousavi450

This does not answer your BiSeq question directly, but I've been very happy using metilene for DMR calling.

ADD REPLYlink written 20 months ago by Chris Miller21k
2
gravatar for Gordon Smyth
7 months ago by
Gordon Smyth980
Australia
Gordon Smyth980 wrote:

edgeR can be used for differential methylation analysis of RRBS data, including chromosomal level tests:

https://f1000research.com/articles/6-2055/v2

edgeR estimates biological variation between replicate samples in a very effective way, and provides full linear model capabilities (multiple groups, batch effects, covariates etc.

It can be used for CpG island based analyses or for analyses of preset genomic regions such as promoter regions.

ADD COMMENTlink modified 7 months ago • written 7 months ago by Gordon Smyth980
1
gravatar for Charles Warden
7 months ago by
Charles Warden7.2k
Duarte, CA
Charles Warden7.2k wrote:

My personal preference is to use methylKit or COHCAP for identifying Differentially Methylated Regions

However, I have tested BiSeq for one or two projects.

I think the short answer can be described as follows:

1) First, I think you need to save your BSraw() object.

I had a slightly different way to do this (from the Bismark coverage files), but I kind of have to give advice in terms of my own experience:

BSraw.obj = readBismark(comp.files, as.character(sample.table$sampleID))

2) At least if you follow the instructions (which should save time at later steps), you are supposed to define clusters of sites to analyze:

BSraw.obj= clusterSites(object = BSraw.obj,
                        groups = var1, perc.samples = min.percent.observed,
                        min.sites = sites.per.island, max.dist = max.cluster.dist)

You don't have to define groups at this step, but you are able to so do (for a two-variable comparison)

3) You'll probably want to define some sort of gene annotation object. I'll let you decide how you would like to go about doing that, but I'll call that GenomicRanges object refGR. You can then annotate your sites with BSraw.obj = subsetByOverlaps(BSraw.obj, refGR). There are some extra steps, but I think the most appropriate thing for me to do in this discussion format is to point out the most relevant BiSeq functions.

4) There is a smoothing function:

predictedMeth = predictMeth(object = BSraw.obj, mc.cores=threads)

5) Then, you should perform your CpG site test:

betaResults = betaRegression(formula = ~var1,
                    link = "probit",
                    object = predictedMeth,
                    type = "BR", mc.cores=threads)

I think this is the step that takes the longest time (if I remember correctly).

6) You can then define differentially methylated regions as follows:

DMRs = findDMRs(betaResults,
                max.dist = max.cluster.dist,
                diff.dir = TRUE)

I think you will want to re-annotate the regions (but I was exporting both site and region results, so I wanted to have those for both analyses).

To be honest, you can probably get the best advice from the developer (and my own experience was to prefer testing other methods before BiSeq). However, I hope this helps!

ADD COMMENTlink modified 7 months ago • written 7 months ago by Charles Warden7.2k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1164 users visited in the last hour