Trinity assembly from many samples

Question

When you combine samples for de-novo transcriptome assembly with Trinity, do you suggest limiting the number of reads for each sample? I had read one of Matthew MacManes' papers awhile back suggesting 40 million reads was a good number for Trinity assemblies of most metazoans.

However, I have 28 samples across 3 treatments that I want to use for DESeq and so, if I follow the advice at BioStars here, that's 28x40 million reads = 1.1 billion reads going into a Trinity assembler... This seems like a troublesome amount. How is it suggested to combine these samples in situations like this?

Answer 1

Assuming these reads are from the same type of thing - and they should be if you're planning to use DESeq[2] on them - then it's fine to combine all the reads into a single Trinity assembly run.

All recent versions of Trinity will perform in silico normalization by default (based on kmer counts), so you shouldn't need to do a manual subsampling of reads. Read normalisation can be done separately; details on that can be found here:

https://github.com/trinityrnaseq/trinityrnaseq/wiki/Trinity-Insilico-Normalization

My vague memory is that the kmer counts are only used to get an idea of the total transcriptome size (i.e. for determining the average transcriptome coverage). If you want to get a more complete assembly, then consider only choosing samples that are most likely to have different overall transcript expression (e.g. different tissue types, treatments, etc.). If you've got 28 samples, you shouldn't need to worry about replicates; more variation is better.