Normalization of RNAseq data and the use of de novo transcriptome

Question

[BioBing]

Hi all,

Hoping that some of the RNAseq experts in here are having some pieces of advice on normalization/proceeding of following data analysis:

The study is about how a stressor is affecting a non-model species (no reference genome/transcriptome available) in terms of differential gene expression. We did a deep sequencing to make a reference transcriptome and sequenced the samples at a "lower depth":

1. Reference transcriptome de novo assembled (Trinity) from reads with a sequencing depth of 300 M (PE 2x150 nt). The statistics (TrinityStats), E50N90, BUSCO analysis, Blast2Go, Detonate (comparison of 3 assemblies - chose the best one) looks good. The reference is made from a non-stressed individual of the non-model organism.

2. Triplicate samples of "negative stress control", "positive stress control" and the "treatment" with a sequencing depth of 25M (PE 2x75nt)

How is the best way to use the reference transcriptome in order to determine differential gene expression of the samples? any tips/tricks on how to normalize?

Thank you!

[biofalconch]

I'm no expert, but you could use your reference transcriptome to map reads of your treatments and obtain counts (kallisto, or you can take the single mapping reads as counts I think), however, you will be missing out on all the isoforms specific to that treatment. You can normalize using edgeR's TMM method ( an explanation here), and I am pretty sure the way from there to determine differential expression is pretty standard (maybe look at edgeR's vignettes?).

PS- Is it E50N90 or E90N50?

[BioBing]

Ops, I meant E90N50 :-)

Thank you! I have considered kallisto as well

[Jake Warner]

The Trinity wiki provides a lot of guidance for exactly what you want to do: https://github.com/trinityrnaseq/trinityrnaseq/wiki/Post-Transcriptome-Assembly-Downstream-Analyses

Since you will be aligning to your transcriptome you will want to rescue multi-mapped reads. The Trinity developers recommend Kallisto, Salmon or RSEM.

As a personal note, my workflow is to map to the assembly using bowtie, estimate abundance using RSEM and then normalization and differential testing using edgeR's TMM method.

Answer 1

[theobroma22]

You can use Rsubread, and this will tie into limma/ edgeR so you can normalize using zoom or limma-trend, depending on the library sizes. Then, test for differential expression. I am also no expert but was able to use this pipeline successfully to do exactly what you are trying to do. Hope this helps.