Question

How to regress phenotype ~ gene + covars in limma?

0

Entering edit mode

anon20955 • 0

@anon20955-20571

Last seen 5.7 years ago

Hi all:

I am wondering how to perform the regressions phenotype ~ gene + covariates in R using limma with microarray data.

I know how to regress gene ~ phenotype + covariates. For example, if phenotype is BMI and covariates is age and sex, I can write:

design = model.matrix(~ bmi + age + sex, data)
fit = eBayes(lmFit(eset, design))
results = topTable(fit, coef='bmi')

However, this is not the same as bmi ~ gene + age + sex.

All advice appreciated!

limma gene expression microarray R • 1.5k views

ADD COMMENT • link updated 5.7 years ago by Gordon Smyth 52k • written 5.7 years ago by anon20955 • 0

score 0 · Answer 1 · 2019-04-18

0

Entering edit mode

Gordon Smyth 52k

@gordon-smyth

Last seen 3 hours ago

WEHI, Melbourne, Australia

You can't do that in limma. limma is designed to regress gene expression on sample characteristics, not the other way around.

ADD COMMENT • link 5.7 years ago Gordon Smyth 52k

0

Entering edit mode

What is the best-practices way to perform the regression that I'm interested in? Simply a for loop over the genes plus lm?

ADD REPLY • link 5.7 years ago anon20955 • 0

0

Entering edit mode

Could you elaborate what you are trying to achieve? Are you looking for the most important genes that may affect the phenotype? Perhaps you would be better off doing Lasso regression for that purpose.

ADD REPLY • link 5.7 years ago mikhael.manurung ▴ 280