edger and continuous variables

Question

[adR]

Hello Dears,

I may ask a quick question on the edger? 1. Does edger support continues variables as DESeq 2 does? 2. Do you think the number of differential expression genes from edger different from DESeq 2?

I just want to compare my result got from DESeq 2 with adger?

Thanks Amare

Answer 1

[russhh]

edgeR definitely supports continuous variables. You will observe differences in the differential expression statistics between DESeq2 and edgeR because they are based on related models, but with different approaches to computing normalisation values.

[ATpoint]

Adding on this, please browse the Bioconductor support forum for edgeR with continuous variables. There are plenty of threads via google. The same goes for DESeq2 vs edgeR, e.g. from the DESeq2 author: https://mikelove.wordpress.com/2016/09/28/deseq2-or-edger/

[adR]

Hi Russhh, Thanks for your comment but I have been searching it since then and couldn't get a single helpful line of statement on how to design continuous variables on adageR. Could you please replay to me if any papers you saw or line of commands if available? Thanks!

[russhh]

(ATpoint made the comment) Just put the continuous variable in your design matrix. As you would in limma, or one of the linear regression models in R. And if the (extensive) documentation isn't sufficient for your needs, politely tell the author. You haven't provided any information about the model you are fitting and haven't provided any code, so there's really nothing we can do to help you at present. Show us what you think you should be able to do (eg, how you defined your model for DESeq2, what happened when you used the same formula for edgeR), any errors that are being thrown up etc and we might be able to help more effectively.

[adR]

Hi Dear AT point, Thank you so much for your regular efforts to answer our question in this platform. Yes, as you commented above I have tried that. Maybe, I may ask your quick look the codes below incase?

---DGEList

FBG <- DGEList(counts = count, group = Sample[,11])

---Design matrix

fastingGlucose <- Sample[,11] design <- model.matrix(~fastingGlucose) # Sample[,11] is a continuous variable

--EstimateDispersion

fbg_est <- estimateDisp(FBG) fbg_et <- exactTest(fbg_est) topTags(fbg_et) summary(decideTests(fbg_et)) # I found 0 up and down regulated genes

however, using the following command lines, I found up and down regulated genes.

fbg_est <- estimateDisp(FBG, design = design) fit <- glmQLFit(fbg_est, design = design, robust = TRUE) qlf_fbg <- glmQLFTest(fit) summary(decideTests(qlf_fbg) # I got some genes up and down regulated

So please one more time could you comment on why the result are different in the exact test and glmQLFTest?

Thank you so much!

[russhh]

Thanks for posting your edgeR code. At no point did you ask why your results would be different using (the wildly inappropriate) edgeR::exactTest and (the completely appropriate) edgeR::glmQLFTest. exactTest is for testing for differences between groups and your glucose measurements do not define a grouping of your samples.

[ATpoint]

DESeq manual:

How can I include a continuous covariate in the design formula?

Continuous covariates can be included in the design formula in exactly the same manner as factorial covariates, and then results for the continuous covariate can be extracted by specifying name. Continuous covariates might make sense in certain experiments, where a constant fold change might be expected for each unit of the covariate. However, in many cases, more meaningful results can be obtained by cutting continuous covariates into a factor defined over a small number of bins (e.g. 3-5). In this way, the average effect of each group is controlled for, regardless of the trend over the continuous covariates. In R, numeric vectors can be converted into factors using the function cut.

See https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#how-can-i-include-a-continuous-covariate-in-the-design-formula