Question

Constrast for Time Course Experiments using Limma

0

Entering edit mode

dl17032000 • 0

@8aa1e163

Last seen 9 months ago

Ireland

Hello! I have an experiment involving the comparison of protein abundance of drug-treated cells with the control group for the course of 1, 2, 3, 4 and 5 hours. As I understand, the "9.6.2 Many time points" section of the Limma user guide fits my experiment design, and I am using the following formula for design:

design <- model.matrix(~0+groups*times)

where groups is:

groups <- as.factor(experimental_design$group) # Either Treatment or CNTR

and times is:

times <- ns(experimental_design$time, df=4)  # Natural splines based on a column with 1/2/3/4/5 hour indicator

Afterwards, I fit the model and its coefficients as described in the guide:

fit <- lmFit(proteins, design)
fit <- eBayes(fit)
topTable(fit, coef=8:12)

However, the code from the user guide does not tell me the log fold changes of those proteins. Thus, I wonder if it is correct to apply the following contrast to my design:

contrast <- makeContrasts(
  (Treatement + Treatment_2 + Treatment_3 + Treatment_4 + Treatment_5)/5 - CNTR,
  levels = colnames(design)
)

and then extract log fold changes from:

fit <- lmFit(proteins, design)
fit2 <- contrasts.fit(fit, contrast)
fit2<- eBayes(fit2)
topTable(fit2)

Many thanks for your time reading this post!

limma • 901 views

ADD COMMENT • link 9 months ago dl17032000 • 0

1

Entering edit mode

Do you have replicate samples for each treatment at each time?

ADD REPLY • link 9 months ago Gordon Smyth 52k

0

Entering edit mode

Yes, I have 6 replicas (samples) for each treatment time (5* 6 = 30 treatment samples in total) and 6 samples for the Control.

ADD REPLY • link 9 months ago dl17032000 • 0

score 2 · Accepted Answer · 2024-03-06

2

Entering edit mode

Gordon Smyth 52k

@gordon-smyth

Last seen 9 hours ago

WEHI, Melbourne, Australia

For your experiment with replication and with 5 time-points there is no need to use splines. It would be simpler and better to follow either Section 9.6.1 or Section 9.5.3 of the limma User's Guide.

Using splines is more complex and has no advantage in your case. The contrast you suggest isn't actually defined for the spline model. For the non-spline model, the logFCs will be available in the usual way.

ADD COMMENT • link 9 months ago Gordon Smyth 52k

0

Entering edit mode

Many thanks for your suggestion! I have followed the 9.6.1 part of the guide and compared the treated groups independently with CNTR.

However, I am struggling with the Section 9.5.3 of User's Guide - with the part where I need to define contrast.

As I understand, my design would be

 model.matrix(~0+groups+groups:times)

so, is it correct to use the following contrast?

contrast <- makeContrasts(
  (Treatment+ Treatment_2 + Treatment_3 + Treatment_4 + Treatment_5) - CNTR, 
  levels = colnames(design)
)

Or would would you advise to follow a different approach for 9.5.3 part?

ADD REPLY • link 9 months ago dl17032000 • 0

0

Entering edit mode

I don't quite understand your question. If you have followed 9.6.1 then you don't need 9.5.3. You should follow one section or the other but not both. They achieve the same ends but with different parametrizations.

If you have compared the treated groups with the control then you've already formed all the necessary contrasts.

ADD REPLY • link 9 months ago Gordon Smyth 52k

0

Entering edit mode

Well, I am just trying to understand if there is a way to find the overall effect of the treatment through all five hours of the experiment because by comparing the individual time points with CNTR, I see that the abundance of different proteins changes inconsistently and, moreover, some of them, for example, decrease at 2 hours but increase at 4 hours of the treatment.

ADD REPLY • link 9 months ago dl17032000 • 0

0

Entering edit mode

You can subset your data to those genes that are significant at any time and then use k-means clustering to group into sets of genes with the same time-dependent pattern.