2
$\begingroup$

I have Illumina RNAseq data and would like to maximize my power to find candidate genes that are differentially expressed genes between experimental conditions.

Many of my (de novo assembled and annotated) gene models have multiple isoforms. Because I will only be using "uniquely" aligning read counts (as specified by DESeq2 documentation) I fear that I might be disadvantaging myself by aligning my reads back to the full set of gene annotations.

Would I be wrong to first filter my gff file for only the longest isoform prior to read alignment?

$\endgroup$
2
  • $\begingroup$ Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. $\endgroup$
    – Community Bot
    Mar 29, 2023 at 23:51
  • $\begingroup$ What program are you using to do read alignment? Please give us the command line (or steps) you used, as the alignment process is different for different programs, and may influence how the transcript models are adjusted. $\endgroup$
    – gringer
    Mar 30, 2023 at 19:36

2 Answers 2

2
$\begingroup$

The fact that an alignment is unique or not is relative to the genome, not its annotation. In other words, a read will be considered a multimapper (and ignored by some programs) if it can align to several genomic positions, but a read will be considered properly mapped if it aligns to a single genomic position, that may or may not correspond to multiple isoforms.

So if you're only interested in gene expression, there shouldn't be any problem with multimappers.

Note that this answer could change depending on what you are aligning on, and how exactly you performed your de novo assembly.

$\endgroup$
1
  • $\begingroup$ Thanks. I think that you are correct about how "unique" is generally defined. I now feel more confident going ahead while keeping alternate isoforms in by gff file. $\endgroup$
    – DavidR
    Mar 31, 2023 at 20:22
1
$\begingroup$

If you want to account for changing abundances of isoforms, use a pseudoaligner like kallisto for counting. And use DEXSeq, not DESeq2, as DESeq2 is intended to be applied to gene counts, not transcript counts.

$\endgroup$
2
  • $\begingroup$ Yes, and just to be clear, I AM doing read counts per gene. My concern was that those counts could get diluted when alternate transcripts of the same gene are present. However, I believe my understanding of "unique" was wrong and Alexlok's answer below seems correct (from my subsequent searching). Nevertheless, your answer has prompted me to also add DEXseq to my pipeline. So thank you! $\endgroup$
    – DavidR
    Mar 31, 2023 at 20:34
  • 1
    $\begingroup$ Small note, if you do want to look at alternative splicing, in my hands DEXSeq on exon bins was quite inaccurate, I got much better from junction-based tools like spladder and MAJIQ; observation which is essentially confirmed by this recent benchmark. If I'm not mistaken, swbarnes2 is recommending DEXSeq at the transcript-level, which I have no experience or opinion about. $\endgroup$
    – Alexlok
    Mar 31, 2023 at 21:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.