samr vs limma packages. How do you decide between them?

Question

[alfonsosaera]

Dear BioStar Users,

I am new in the BioStar community, I tried to search for posts that answer my question but I couldn't find any. Please forgive me if this has been already answered.

I need to analyze a microarray dataset so I decided to use the bioconductor package I knew for this task, limma. Limma uses linear models to find Differentially Expressed Genes (DEGs) between 2 groups that provides p-values and q-values.

However, a colleague of mine suggested me to use SAM: Significance Analysis of Microarrays available for R in CRAN as samr. SAM uses a non-parametric approach to find DEGs based on a d score (reflecting the difference of the means of the 2 groups) and a FDR.

My question is, how do you decide between them? which one do you think is "better"?

A related question is, if you decide to use SAM, how do you prepare exploratory graphs such as volcano plots?

Thanks for your hep!

Alfonso

[Benn]

I am not sure if you can get an answer to which is "better", but what I can tell you is that limma is by far the most popular tool used for microarray DE analysis, and that it is very well accepted as the standard tool to use.

[alfonsosaera]

Thanks for your comment @b.nota, I started with Limma because it is the accepted standard for microarrays but then heard about samr and I could not find any benchmark or comparison of the 2 methods.

[Benn]

Maybe a pubmed search for "limma sam" can help? e.g. https://www.ncbi.nlm.nih.gov/pubmed/26057385

[alfonsosaera]

Thanks again! the paper is very interesting. From the paper it appears that Limma and SAM work very similar when DEGs differ hihgly from the non-DEGs in the dataset. When the DEGs are closer to the non-DEGs Limma performs slightly better.

Answer 1

[Charles Warden]

I have to admit I typically use limma over samr.

However, to be fair, there probably isn't one absolute "best" option. If you are having difficulty with limma or a standard statistical test (while doing things like changing FDR and |fold-change| criteria), perhaps it is worth trying out samr.

[alfonsosaera]

Thanks for your answer! I think I will stay with limma, it is more or less accepted as the standard and it provides p-values and q-values (FDR) that fit most of my needs.