What was the very first paper to describe "limma-trend"? By "limma-trend" I mean the one described in Law (2014), but I am not sure that was the first time.

Thanks,

Nik

What was the very first paper to describe "limma-trend"?

1

Entering edit mode

What was the very first paper to describe "limma-trend"? By "limma-trend" I mean the one described in Law (2014), but I am not sure that was the first time.

Thanks,

Nik

1

Entering edit mode

Based on the citations from this paper: http://nar.oxfordjournals.org/content/43/7/e47

I believe it is this one:

"Intensity-based hierarchical Bayes method improves testing for differentially expressed genes in microarray experiments." http://www.ncbi.nlm.nih.gov/pubmed/17177995?dopt=Abstract

(Edit: As pointed out, this paper describes a method for analyzing microarrays. If you're looking for the RNA-seq analysis method, Gordon's answer is correct.)

1

Entering edit mode

The article by Law et al (2014)

http://www.genomebiology.com/2014/15/2/R29

proposed the "limma-trend" method for analysing RNA-seq data. It originated both the voom and limma-trend methods. The article says:

*"We explore two ways to incorporate the mean-variance relationship into the differential expression analysis. ...We call the first method limma-trend and the second method voom"*

The limma-trend method consists of computing logCPM values using edgeR, then analyzing these in limma using trend=TRUE in the eBayes call. The method is shown to perform as well as the "voom" method except when the library sizes are very unequal.

A Google Scholar search shows that this is only published article to have used the term "limma-trend".

1

Entering edit mode

In our article (Law et al, 2014), we did use edgeR to compute the logCPM values as part of the limma-trend algorithm. Specifically we used edgeR's cpm() function with a prior count of 1. You can confirm that from the R code provided in supplementary materials. Sorry if that was not explained in the main text.

The CPM are not a very hard calculation so you could easily compute them directly. In practice, we frequently use TMM normalization in conjunction with limma-trend, just as we do with voom, so we need edgeR anyway. We used edgeR logCPM instead of voom logCPM for reasons explained here:

Differences between limma voom E values and edgeR cpm values?

As I've already told you, there was no earlier article. Why do you keep asking? Our article (Law et al, 2014) tells you that the method is being proposed for the first time. If there had been an earlier article, then we obviously would have cited it.

I was senior author of the Law et al paper, and I was the one who created and coded the limma-trend method, so I'm obviously in a good position to tell you about its origins.

Similar Posts

Loading Similar Posts

Traffic: 465 users visited in the last hour

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

Maureen Sartor's article was the first to suggest that limma's eBayes algorithm be extended to include a trend on the prior. Although the suggestion was made in the context of microarrays, the idea certainly contributes to the limma-trend algorithm for RNA-seq analysis. Law et al (2014) cites the above article for that reason.