Can one get LSMeans and expression confidence intervals from limma and voom?
Entering edit mode
bag59 • 0
Last seen 5.4 years ago


I am processing transcriptome-wide expression data using the limma pipeline. I have a two factor experiment (2 treatments crossed). My data are standardized with voom, precision weights are calculated with eBayes, and analysis of differential expression is done though limma modeling.

Does anyone know if an LSMean value and confidence intervals for that expression can be obtained from these tools? Limma offers options for outputting CI's on the log fold change between treatments and average expression across all treatment group combinations,  but I have yet to find a way to output CI's or LSMean values for the straight-forward normalized expression values of a particular gene in a particular treatment combination.

The R package lsmeans does not appear to work with MArray model objects generated by limma. Voom and eBayes offer options for outputting precision weight values for each gene, but I am unsure how to transform this information into confidence intervals.

I am primarily interested in this information for plotting purposes, for data on individual genes.

Any help appreciated. Thanks in advance.





ebayes limma voom • 1.0k views
Entering edit mode
Aaron Lun ★ 27k
Last seen 2 hours ago
The city by the bay

If you have two crossed treatments, it should be easy enough to express your experimental design as a one-way layout containing four groups (i.e., one group per combination of factors across the two treatments). If you fit a linear model using this design matrix, the values of the coefficients will be the group means. You should also be able to drop each coefficient with topTable; the "log-fold change" here would be the mean (log-)expression in the corresponding group, with the corresponding confidence intervals obtainable by setting confint.

P.S. Separate limma and voom in your tags, otherwise the maintainers don't get informed.

Entering edit mode

Excellent! Thanks Aaron!! This worked perfectly, although it took me a few tries to get the correct design matrix for the output I wanted. I had to be sure and use the group-means parameterization from the User Guide. I didn't know the "logFC" output corresponds to the level means when there is only one factor in the model. That's a useful tidbit.


Login before adding your answer.

Traffic: 349 users visited in the last hour
Help About
Access RSS

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6