Using Salmon to map RNA-Seq and Ribosome Profiling reads in Yeast and downstream analysis with DESeq2

Question

[kyusikkim]

Hello all! I was just wondering what the best method might be to use Salmon to map reads for both ribosome profiling and RNA-Seq and analyze the results using DESeq2. I have built a transcriptome with UTRs and used Salmon to map my reads. However, I want to only count reads that map to the CDS and not UTRs in the ribosome profiling dataset. Generating two different transcriptome indexes leads to problems with downstream analysis with DESeq2. For the profiling reads, would it be advisable to map against the transcriptome using an aligner such as Hisat2/STAR, manipulate the bam files to remove reads that align to UTRs, then quantify using Salmon? Would this lead to inaccurate quantification? If not, is there some way to use DESeq2 to analyze featureCounts quantified ribosome profiling reads against Salmon mapped RNA-Seq reads?

Answer 1

[chasem]

My suggestion would be to align the reads using your favorite aligner. I use the nf-core RNAseq pipeline, which would also give you the option of STAR/Salmon, RSEM and I think another, for what it is worth. If you do use STAR or another splice aware aligner, I would suggest setting the intron max to the maximum intron size in your transcriptome.

For the CDS quantification, if I were going to do this I would use the bam files produced by the alignment step above, and then use HTseq or FeatureCounts as you said. For both programs, you'll set the feature type over which the program counts to CDS. In HTSeq, the option is -t. This is under the assumption that you're looking at gene expression, and not transcript expression, of course.

I don't personally see why you couldn't compare the transcript quantification aggregated to the gene against the HTSeq/FeatureCounts gene counts using the CDS as features at this point -- the question would be how much does including the UTR affect gene expression. That said, you could use HTSeq or FeatureCounts to count over the exons to include the UTR and eliminate Salmon, etc -- that would arguably make them more comparable. I realize that HTSeq and FeatureCounts are less commonly used for quantification now, but for yeast especially I think they're reasonable methods.

Another method would be to count expression over the UTRs directly -- HTSeq and FeatureCount could do that, too.

[kyusikkim]

That's a good point. I think making the custom annotation, mapping to the genome using STAR, and then using HTSeq for both the RNA-seq and the ribosome profiling might be the way to go.