RNA-seq analysis with quantitative traits

Question

[javiergutierreza]

Dear Biostars: Much of the DGE analyses tutorials and examples are made using binary traits (Presence-Absence) (or a "binarization" of a trait). However, there is not much about dealing with traits in the quantitative scale (Please correct me if I'm wrong).

The basic design of my RNA-seq experiment in a binary fashion will be:

Group Sample TimePoint

A Sample1 0

A Sample1 1

A Sample2 0

A Sample2 1

B Sample1 0

B Sample1 1

B Sample2 0

B Sample2 1

and the design I'm using is = ~ Group + Group:Sample + Group:TimePoint

My question is: Could be possible to run the same analysis in a Quantitative continous fashion? Example matrix:

Group Sample Density

A Sample1 1.3

A Sample1 5.5

A Sample2 1.2

A Sample2 6.7

B Sample1 0.8

B Sample1 3.4

B Sample2 1.6

B Sample2 7.1

and use the same design matrix for the analysis? : ~ Group + Group:Sample + Group:Density

Could this analysis be done using standard DGE analysis packages like DESeq2, edgeR or limma-voom? or will be better to perform linear regressions (on it different tastes) with the quantitative trait as response and normalized gene expression as the explanatory variable?

Any help/thoughts will be much appreciated!!

Best,

Javier

Answer 1

[Devon Ryan]

Sure, you can use a quantitative value like that in DESeq2, edgeR or limma/voom.

[javiergutierreza]

Hi Devon, Thanks for your answer. It is not necessary to add any tweaks, compared to the usual analysis with binary traits?

[Devon Ryan]

Nope.

Answer 2

[TriS]

this is also a good read for your analysis

RNA-seq analysis for detecting quantitative trait-associated genes

[javiergutierreza]

Hi TriS, Indeed, that was the first (and only) paper dealing with this kind of analysis that I have found so far. It's helpful, however, is a bit confusing for me, because they propose to use the quantitative trait as the response and the gene expression as an explanatory variable. What I'm not sure is if DESeq2, edgeR and/or limma-voom will take this into account.

Answer 3

[jnoble333]

Hey Javier,

There isn't much work that has been done regressing the transcriptome onto quantitative phenotypic traits. I am attempting to do this as a component of my thesis. Here's a little insight I can provide:

Differential expression packages like EdgeR, DeSeq, et al. will not be suitable for what you are trying to achieve if you have traits measured in a population. You will likely need some for of least squares regression with a likelihood testing component. There has been work done looking at the correlation between gene expression and phenotypic traits however (eg https://genomebiology.biomedcentral.com/articles/10.1186/s13059-015-0735-9). I've run across a few papers that integrate SNP and gene expression data to discover predictors of phenotypic traits. One of these is http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0115532#s3 and they use the BLR package from the de los Campos group. The newer version of this package is BGLR and Dr. de los Campos mentioned using it to regress gene expression on phenotypic data at PAG but I haven't seen and specific examples of this.

Hope some of this helps,

JD

[javiergutierreza]

Hey JD, Thanks a lot for your insights and the papers. I though this kind of analysis was more common, but seem like not.