How to understand logics and statistical concepts behind limma package?

Question

[Sib]

I want to prepare a tutorial on the logic and statistical concepts behind differential expression analysis of the limma package. My audience is biologists. So the tutorial I prepare should explain that complicated statistical concepts in simple words. What I have in my head is like the StatQuest videos. Actually, I should explain in the tutorial, this limma package's article that reviews the philosophy and design of the limma package: https://academic.oup.com/nar/article/43/7/e47/2414268 .

But I am a biologist and understanding this article is hard for me. Would you please give me a list of concepts that I should learn in order so that I can understand this article? Also please introduce to me some sources for learning them.

Answer 1

[i.sudbery]

As noted by @papyrus, until you understand linear models, you are going to struggle to understand limma. So start there. If you don't get on with any of the links mentioned by papyrus, you could alsod try: http://rafalab.github.io/pages/harvardx.html

The key technical advance of limma was the use of empircal bayes to produce accurate estimates of variance, even when only a small number of replicates was available. This is what originally made limma better than t-tests, or R's standard lm linear modeling (it now has a whole load of extra bells and whistle that do all sorts of things not possible before).

The best explaination of empirical bayes for me was a series of blog posts by David Robinson on the VarianceExplained blog. I'd probably start with this one: http://varianceexplained.org/r/empirical_bayes_baseball/

These have now been released as a ebook, which is pay-what-you-want: http://varianceexplained.org/r/empirical-bayes-book/

[Papyrus]

Wow, those posts/ebook by David Robinson are great!! thanks for sharing!

[Sib]

Hi @i.sudbery. Thanks for the resources you introduced. The videos of the first source are private on youtube. What is that for? Is there a way that I can watch them?

[i.sudbery]

Oh sorry. That didn't used to be the case. They are video lectures from an EdX course. You can find a link here:http://rafalab.github.io/pages/teaching.html

Apparently there is a free audit option for all the courses.

[GenoMax]

There is a line on page that says:

Note that you must be logged in to EdX to access the course.

EdX accounts are free or so says the page.

Answer 2

[Papyrus]

IMHO, the most important thing before getting into the rest of the (numerous) technical possibilities of the package, is to have a good knowledge of linear regression. For example through courses such as this one. Linear regression is the basis of the strategy used by limma and other packages to model experiments. Another resource I'm particularly fond of is the series of Points of Significance which summarize statistical concepts for a biological audience.

Answer 3

[Sib]

I found an excellent source for understanding all the statistical concepts of differential analysis of microarray with all the perquisites. This is an excellent course from PennState Eberly College of Science. I put that link here so that others can use it if they had the same issue:

https://online.stat.psu.edu/stat555/node/1/