remove common unmethylated CpGs about bsseq HOT 6 CLOSED

On Wed, Jun 16, 2021 at 8:23 AM RRebo ***@***.***> wrote: Yes of course, I should have said in my original post. I am working with insects so the methylation is highly concentrated in CpG methylated regions, while the whole genome is unmethylated. A few "insect methylomes" are analyzed by removing common unmethylated CpGs (if I understood correctly) in order to increase the detection of DMRs. Here is a paper summarizing it: https://europepmc.org/article/pmc/pmc6357555 We thus suggest removing globally unmethylated CpGs before proceeding with statistical analyses.[...] We then used Bis-class [31] to identify and discard sites that were globally unmethylated in all 12 samples. We also removed CpG sites that did not have any mapped cytosine reads in any samples. We recommend using small window sizes for smoothing, removing globally unmethylated CpG sites before DML/DMR analyses and applying a more specific categorical classification method. These considerations should be applicable to a variety of species that share similar methylation characteristics with invertebrate genomes. What do you think? (and thank you) Rita — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#102 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABF2DH2ICTPAZNKOL4GYFSDTTA7NFANCNFSM46XTTY6Q> .

I have worked a bit on insect methylomes in the long past (bee, Herb et al 2011). As you're saying - at least with my limited experience - the methylation structure (if they have methylation of course) is radically different from mammals. The thing you're quoting probably suggests to remove unmethylated CpGs to decrease the multiple testing burden, which makes sense to me, although it is critically dependent on how you actually identify DMRs. Smoothing with a narrow bandwidth also makes sense to me based on the methylation patterns (although I don't know if bee's are general). Although it must be said that the more narrow the bandwidth, the less advantage you might get with smoothing. However, removing unmethylated CpGs _prior_ to smoothing is a bit weird to me. Usually I think of smoothing as estimating a local "regional" methylation rate. If you remove all unmethylated CpGs prior to smoothing, you will get much higher methylation estimates at a regional level. Now, one might be able to argue for it because of the pattern, but it really changes the interpretation. Removing CpGs prior to smoothing is easy. You just subset the BSseq object before you run BSmooth. Best, Kasper On Wed, Jun 16, 2021 at 8:23 AM RRebo ***@***.***> wrote: > Yes of course, I should have said in my original post. I am working with > insects so the methylation is highly concentrated in CpG methylated > regions, while the whole genome is unmethylated. A few "insect methylomes" > are analyzed by removing common unmethylated CpGs (if I understood > correctly) in order to increase the detection of DMRs. Here is a paper > summarizing it: > https://europepmc.org/article/pmc/pmc6357555 > > We thus suggest removing globally unmethylated CpGs before proceeding > with statistical analyses.[...] We then used Bis-class [31] to identify and > discard sites that were globally unmethylated in all 12 samples. We also > removed CpG sites that did not have any mapped cytosine reads in any > samples. > > We recommend using small window sizes for smoothing, removing globally > unmethylated CpG sites before DML/DMR analyses and applying a more specific > categorical classification method. These considerations should be > applicable to a variety of species that share similar methylation > characteristics with invertebrate genomes. > > What do you think? (and thank you) > > Rita > > — > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub > <#102 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ABF2DH2ICTPAZNKOL4GYFSDTTA7NFANCNFSM46XTTY6Q> > . > -- Best, Kasper

On Wed, Jun 23, 2021 at 9:04 PM RRebo ***@***.***> wrote: Thank you very much. I'll try both ways (to keep the common un methylated and to remove them) and check the results! For now I am stuck in the tstat because it gives me a "need at least two non-NA values to interpolate" which I read it might be due to small contigs (I have tones of that!). I'm trying to figure out how to remove them :) So thank you again and I guess we can close the issue. Rita — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#102 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABF2DH76DVIDONBN6O6XIULTUIV3DANCNFSM46XTTY6Q> .