Question

limma interaction term

0

Entering edit mode

PJ • 0

@pj-8012

Last seen 7.2 years ago

United States

Hi everybody,

I have a model with 2 main factors and its interaction:

designmodel = model.matrix(~0 + Factor1 + Factor2 + Factor1:Factor2)

Factor 1 has two levels and factor 2 has 4 levels. My question is if when I test the interaction term I can add all 8 groups as follow:

contrastInteraction = makeContrasts(((Factor1.Level1Factor2 - Factor1.Level2Factor2 - Factor1.Level3Factor2 - Factor1.Level4Factor2 ) - (Factor2.Level1Factor2 - Factor2.Level2Factor2 - Factor2.Level3Factor2 - Factor2.Level4Factor2)), levels=design)

Or shoud I test the interaction in 4 separate contrasts (one for each level of factor 2)? Thanks for you advice!

PJ

limma interaction model • 1.4k views

ADD COMMENT • link updated 7.8 years ago by Gordon Smyth 50k • written 7.8 years ago by PJ • 0

1

Entering edit mode

You don't provide a lot of details - what's your experimental design? What does Factor1 or Factor2 contain? - so it's hard to tell what the coefficient names mean, let alone whether you're doing something sensible with your contrasts.

ADD REPLY • link 7.8 years ago Aaron Lun ★ 28k

score 0 · Answer 1 · 2016-06-25

0

Entering edit mode

Gordon Smyth 50k

@gordon-smyth

Last seen 55 minutes ago

WEHI, Melbourne, Australia

You model has 3 degrees of freedom for interaction. To test for interaction you would do:

fit <- lmFit(y, design)
fit <- eBayes(fit)
topTable(fit, coef=6:8)

There is no need to use makeContrasts().

Note that you must test all 3 df together. Trying to do separate tests for individual interaction contrasts has no meaning.

ADD COMMENT • link 7.8 years ago Gordon Smyth 50k