Interaction term analysis using ballgown
Entering edit mode
jcalis ▴ 10
Last seen 5.3 years ago

Dear all,
I am trying to analyze a Stringtie based gene expression data set using ballgown, in a way similar to what was described for the analysis of difference of differences or interaction terms using voom+limma (see 9.5 in
It would be nice if I can make the same analysis using stringtie+ballgown using FPKM values. 
Let me describe the setup, there are 18 samples for each combination of these characteristics:
- genotype: A, B or C
- treatment: plus or minus
- replicate: 1, 2 or 3
I am trying to find genes/transcripts where the response to treatment is different for different genotypes. For instance, a gene that would be upregulated in versus A.minus, but not in versus B.minus. Is it possible to make this analysis using the nested models approach, or in some other way?

Can anybody show an example of this analysis using ballgown, or point me in the right direction?


ballgown stringtie voom limma differential gene expression rnaseq • 1.5k views
Entering edit mode
Alyssa Frazee ▴ 210
Last seen 3.7 years ago
San Francisco, CA, USA

This should be possible using the nested model approach in ballgown. In fact, you can use the same nested model structure you might use with limma (ballgown's "stattest" has arguments where you can specify your own models). I'd imagine you'd specify your smaller model as intercept + genotype + treatment, and the larger model as intercept + genotype + treatment + treatment*genotype (the interaction term), which would basically give you genes where the interaction term improved the expression prediction, i.e., the genes where the treatment effect differs by genotype (which I believe is what you're looking for).

It might be better in this case to use limma rather than the built-in models with ballgown -- you can still use the ballgown functions to load, plot, and explore your data, and to get everything in the correct matrix form needed by limma, but the smoothing / sharing of information across genes provided by limma might be useful in this context since you have only 3 replicates for each genotype / treatment combination.

(Also, may be helpful to hear from someone with more recent experience doing an interaction analysis like this! I can speak to ballgown's capabilities, but haven't done a gene expression analysis in a while, so other perspectives could also be useful here.)

Entering edit mode

Thanks! I'll look into the models-approach and see how it compares with a limma-based analysis. 

Entering edit mode

Hi jcalis,

I am trying to run a differential expression analysis using ballgown, also with an interaction term, and I'm curious to know what has happened with your attempt.

Any insight would be greatly appreciated.



Login before adding your answer.

Traffic: 449 users visited in the last hour
Help About
Access RSS

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

Powered by the version 2.3.6