Three way comparison RNAseq libraries
1
1
Entering edit mode
9.5 years ago

Hello,

I have 3 RNAseq libraries samples with 3 replicates each. I wanted to know what are the RNAseq analysis options for doing a three way comparison?

I've seen that DEseq can do differential gene expression analysis for two or more samples.

I am also interested in differential isoform expression, but Tophat/Cufflinks only does two way comparisons for RNAseq libraries. Will I have to do 3 two-by-two comparisons from Tophat/Cufflinks to do a three way comparison from Tophat/Cufflinks outputs?

Or is there an easy method to do three way comparison of RNAseq libraries?

Thank you

Rodrigo

RNA-Seq R Analysis • 4.3k views
ADD COMMENT
2
Entering edit mode
9.5 years ago
Steven Lakin ★ 1.8k

I believe the answer to this depends on your experimental design. By definition, a differential expression analysis is a comparison between two states, usually that you expect to have different biological phenotypes/outcomes. If all three of your libraries represent different experimental conditions that you suspect are independent against the dependent library, then the best way to analyze them is pairwise.

If you have a situation where one effect on one dependent condition is due to multiple other independent conditions, then you can use generalized linear models, which is supported by DESeq. For example, if you suspect your differential result is a result of changes in both your treatment and in the sequencing platform used (let's say you used Illumina to sequence one library and Ion Torrent to sequence the others), then multi-factor analysis would be appropriate:

Differentially Expressed Counts ~ Treatment + Sequencing Type

For my RNA-seq DE analyses, I've done pairwise comparisons for the most part, using DESeq, baySeq, and edgeR, and I looked at: 1. the output of each package, 2. the intersection of any two of the three packages, and 3. the intersection of all three packages. You can get a sense of which of your DE calls are more statistically significant this way. Often, it is worth using more than one method, since they all preprocess and analyze the data in different ways.

So yes, you'll likely end up doing 3 pairwise comparisons for these, unless your experimental design is built for GLMs.

ADD COMMENT

Login before adding your answer.

Traffic: 2438 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

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

Powered by the version 2.3.6