Question

edgeR supports continuous covariates?

0

Entering edit mode

Ivy • 0

@user-24077

Last seen 3.9 years ago

Thanks for preparing the guideline on edgeR. However, I did not see any information supporting continuous variables in edgeR. May I know does edgeR supporting below 2 situations:

(1) 2 group comparison with adjustment of1 or 2 continuous covariates? (2) Association analysis of gene with one continuous variable? Thanks!

edgeR • 2.4k views

ADD COMMENT • link updated 22 months ago by Jeffcgen2000 • 0 • written 4.1 years ago by Ivy • 0

score 3 · Answer 1 · 2020-11-06

3

Entering edit mode

Yunshun Chen ▴ 900

@yunshun-chen-5451

Last seen 8 days ago

Australia

edgeR can handle time-course experiments where time is considered as a continuous variable. Please refer to the time course case study in the user's guide. The cubic spline approach can be applied to continuous variables other than time.

ADD COMMENT • link 4.1 years ago Yunshun Chen ▴ 900

0

Entering edit mode

Hi Yunshun Chen and Gordon Smyth,

Thanks for the quick response. It would be highly appreciated if you can explain a little bit further.

I see poly(Hours, degree = 3) and ns(Hours, df =3) before constructing the design matrix in the user's guide. Are there any reason in using this transformation of continuous covariate? Why cubic spline is suggested to continuous variable? I am confused in which df can be used in ns().

Furthermore, as Gordon Smyth suggested here "edgeR can handle any linear model with any combination of continuous covariates or factors. Any model that can be created by model.matrix can be fitted by edgeR."

Is that ok to use the following code directly to both factor covariate and the continuous covariate analysis? Much appreciated!

d <- edgeR::calcNormFactors(d) 
design <- model.matrix(~covariate)
rownames(design) <- colnames(d)
d <- estimateDisp(d, design, robust=TRUE)
fit <- glmFit(d,design)
res <- glmLRT(fit)

ADD REPLY • link 4.1 years ago Ivy • 0

score 2 · Answer 2 · 2020-11-06

2

Entering edit mode

Gordon Smyth 52k

@gordon-smyth

Last seen 27 minutes ago

WEHI, Melbourne, Australia

edgeR can handle any linear model with any combination of continuous covariates or factors. Any model that can be created by model.matrix can be fitted by edgeR.

ADD COMMENT • link 4.1 years ago Gordon Smyth 52k

0

Entering edit mode

Hi Yunshun Chen and Gordon Smyth, does edgeR support mixed model? (https://en.wikipedia.org/wiki/Mixed_model)

ADD REPLY • link 22 months ago Jeffcgen2000 • 0