Question

How is the best way to construct the model.matrix and the lmfit comparing longitudinal data?

0

Entering edit mode

leticia.n.spindola • 0

@leticianspindola-15317

Last seen 6.7 years ago

Hi all,

I’m working with longitudinal data but I am not sure which is the best way to analyse this type of data using limma.

I have the blood methylation data from 24 individuals evaluated at the baseline and after 3 years of follow up. All 24 subjects were healthy at the baseline and all developed the disease after the follow-up.

I would like to compare the methylation values between these two time-points. However, I don’t know how to construct the model considering that I’m comparing the same individuals in two different moments.

I have two variables:

1) Time (baseline or follow-up);

2) Subjects (identifying the different measures for the same sample)

How is the best way to construct the model.matrix and the lmfit?

Any help would be greatly appreciated.

Thanks!

limma longitudinal data limma paired analysis • 1.3k views

ADD COMMENT • link 6.7 years ago leticia.n.spindola • 0

score 1 · Answer 1 · 2018-03-22

1

Entering edit mode

Ryan C. Thompson ★ 7.9k

@ryan-c-thompson-5618

Last seen 6 weeks ago

Icahn School of Medicine at Mount Sinai…

This sounds like a standard paired test design, for which you would generally use a design formula of ~Time + Subject. There are several examples of such designs in the limma User's Guide.

ADD COMMENT • link 6.7 years ago Ryan C. Thompson ★ 7.9k

score 0 · Answer 2 · 2018-03-22

0

Entering edit mode

leticia.n.spindola • 0

@leticianspindola-15317

Last seen 6.7 years ago

Thank you for the answer Ryan. I was wondering if it was necessary to use the function duplicateCorrelation(), since I'm investigating the same subjects and I thought that the comparison should be within subject. In which situation should I use this function?

ADD COMMENT • link 6.7 years ago leticia.n.spindola • 0

1

Entering edit mode

Firstly, reply to answers using "add comment", not "add answer", unless you're answering your own question.

Secondly, duplicateCorrelation() is intended for situations where you want to compare across blocking levels, e.g., across subjects (presumably in different conditions). In such cases, blocking in the design matrix would not be appropriate, as the block effect would be indistinguishable from the effect of condition. Of course, this discussion is mostly academic because in your case, you are comparing within subjects, so Ryan's suggestion is perfectly suitable.