What information in dispersion plot using edgeR?

Question

[ghmdsr]

I'm doing DEG analysis using edgeR. I have some questions.

1. Difference between 'edgeR' and 'Limma'
2. Interpreting 'dispersion plot' : after estimating dispersions

Code:

dgList2 <- estimateGLMCommonDisp(dgList1,design = designMat)
dgList3 <- estimateGLMTrendedDisp(dgList2, design=designMat)
dgList4 <- estimateGLMTagwiseDisp(dgList3, design=designMat)

plotBCV(dgList4)

In this plot, what information can I get? Please help me.

[Kevin Blighe]

Probably better on the Bioconductor forum.

Answer 1

[h.mon]

1. edgeR is intended for RNAseq data, and fits a negative binomial model to test for differential gene expression; limma is intended for microarrays, and fits a linear model to test for differential gene expression.

2. The dispersion is a parameter for the negative binomial model. edgeR uses some fancy techniques to estimate (and possibly squeeze) dispersion from few samples. plotBCV() shows these estimates against the log(counts per million), which is useful for evaluating if the model fit is good and if there is suspicious data lurking in your samples.

[Kristoffer Vitting-Seerup]

I do feel we need to mention that although limma was originally made for microarrays it continues to be amongst the top tools in benchmarks of DE analysis on RNAseq data - especially when using voom to add weights to the data.