Some queries on DNA methylation when processing the BS-seq data

[zeam]

I saw some papers on DNA methylation using BS-seq,now I meet some questions on DNA methylation when working with the BS-seq data:

1)first,what is the minimum required threshold of percent methylation at a site to dedine a methylcytosine,I want to use 40%,but I don't know if it was appropriate;

2)second,somebody mentioned methylation genes in some papers,so how to define a methylated genes,what the criteria is ? Because you know,the raw results of BS-seq is the methylation level at each site,and both of the methylation sites and the C are not even between a gene;

3)Also,I want to know how to define a differentially methylated region(DMR),this question is in fact the same as my second one because of the uneven methylation cytosine in the chromosomes;and in plant how long (300 bp?) to set the length of DMR in your experiences

Thanks for your time!
Best wishes,

[simonandrews]

I don't think you're going to find definitive answers to your questions because a lot of this is going to vary depending on the exact system you're working with.

From the data we've seen the majority of cytosines we measured had a methylation percentage either above 80% or below 20%. Our standard filters are therefore 75% and 25%. However we've also gone back to look at regions with a known methylation state and have found that in those regions even if you apply no filtering at all (so set your cutoff at exactly 50%) the number of incorrect calls we make is still very low (below 5%). The errors also tend not to be clustered so the chance of miscalling a region containing several Cs is even lower.

In terms of identifying DMRs you can either take a purely statistical approach where you look at the proportion of meth vs unmeth in two samples and see if those two groups are significantly different, or you set a cutoff on the amount of change you want to see and then test only regions which pass the initial filter. We took the second approach, which seemed to work out OK. Ideally you don't want to define a size of DMR since they could be of variable length. We've not seen convincing DMRs which were very short though, so you could set a lower cutoff of a couple of hundred bases to remove noise from your results.

[zeam]

To simonandrews

I don't think you're going to find definitive answers to your questions because a lot of this is going to vary depending on the exact system you're working with.

From the data we've seen the majority of cytosines we measured had a methylation percentage either above 80% or below 20%. Our standard filters are therefore 75% and 25%. However we've also gone back to look at regions with a known methylation state and have found that in those regions even if you apply no filtering at all (so set your cutoff at exactly 50%) the number of incorrect calls we make is still very low (below 5%). The errors also tend not to be clustered so the chance of miscalling a region containing several Cs is even lower.

In terms of identifying DMRs you can either take a purely statistical approach where you look at the proportion of meth vs unmeth in two samples and see if those two groups are significantly different, or you set a cutoff on the amount of change you want to see and then test only regions which pass the initial filter. We took the second approach, which seemed to work out OK. Ideally you don't want to define a size of DMR since they could be of variable length. We've not seen convincing DMRs which were very short though, so you could set a lower cutoff of a couple of hundred bases to remove noise from your results.

Hi,there!
Thanks for you reply,and I have one question you didn't answer:how to define a methylated gene,or how to define a methylated region.Imaging I have the methylation information (0-1) at each context(CpG,CHG,CHH )for each chromosomes,how to define a methylated region(what the algoritm is) in your experience?
Thanks again!

[sanamjeet]

Methylation level/percentage

I want to ask what does it mean by methylation level ? is it the percentage of the sequence in whole genome methylated ? or something else.

[simonandrews]

I want to ask what does it mean by methylation level ? is it the percentage of the sequence in whole genome methylated ? or something else.

It will be the percentage of all cytosines which are methylated. In some cases the cytosines will be divided into different contexts (CG CHG CHH etc) and you can quote a methylation level for each context.