Reference for Salmon

Question

[jh]

Hi,

With Salmon, reads are mapped against the transcriptome and not the genome. Is it possible to use an extracted transcriptome from the human genome prepared for RSEM?

Thanks!

Answer 1

[Rob]

Yes, if you prepare a reference with rsem-prepare-reference it will work with salmon.

[jh]

Rob: Which transcriptome reference file do you recommend I build the index from? And where can I get such files? :) I want to analyze two different outputs: (1) isoforms of known genes in the human genome, and (1) total transcript abundance of known genes in the human genome with HGNC gene names.

[Rob]

I generally like the GenCode reference, since it's fairly comprehensive and you can get the sequences directly (rather than having to build them from the genome + gtf). You can also get just the protein-coding transcripts if you want. I generally shy away from e.g. the complete Ensembl transcriptome since it includes many exact duplicate sequences (with different names), that should generally be filtered out or collapsed before quantification.

[jh]

Thank you so much!! :D

[azlanazali01]

Some papers used RSEM for transposable elements/repeats quantification. Does this mean I can use the rsem-prepare-reference of the repeat library, and feed it into Salmon?

[Rob]

Users have had success using salmon for such quantification tasks before. So, if you provide it with the proper reference, yes; it should work as well as RSEM for this task.

[ponganta]

Hi, I have a question: does the result consist of predicted CDS? Because I was wondering whether you can use predicted CDS instead of e.g. assembled transcripts with Salmon. I encountered some chimeric transcripts during de novo assembly, and expect a large amount of collapsed homeologs, too (polyploid species). Transdecoder was able to disentangle concatenated CDS.