DSS "smoothin.span" parameter

[joao-chrusciel]

Hi!! I’ve been working with 6mA data and Nanopore sequencing recently and am trying to perform differential 6mA analysis. I noticed that you used a smoothing.span of 150bp in your pipeline with the DSS package. Is there a specific reason for choosing this value?

The DSS documentation mentions a default span of 500bp, but I couldn’t find directions on how much this value can be reduced or what it depends on. Could you share your insights on the 150bp value you chose?

[vahidAK]

Hi,
The 500 bp in DSS is for CG methylation in WGBS. Adenines are denser than CGs so I used a smaller number, also 150 bp worked well in my experiments.
Best,
Vahid

[joao-chrusciel]

Thanks for the fast answer. If you don't mind me asking, when you say it worked well within your experiments do you mean that you could see some interesting outputs in DMLtest?

The main thing is that when I try a span of 150bp I get almost identical outputs (FDR, pval, mu1, mu2, diff and stat) for many positions which seems odd, but when I decrease it to like 50bp I get interesting and more variable values (which seems to work better). To my understanding, smoothing would increase the accuracy of mean estimates, but I'm not sure if decreasing it too much would mean that I would be incrementing noise in the data or lowering accuracy. Also, I'm dealing with 4 controls x 4 conditions, so I'm not sure if smoothing is truly necessary.

Best,
João

[vahidAK]

I meant DMRs and not DMLs. For DMLs you kind of expect that because you smooth the As in that window and DML is the results for each differential A site. You should look into DMRs that show differentially methylated regions you are looking for.
Yeah, if you have several controls and cases, the smoothing is not necessary but I guess recommended.

Best,
Vahid