Isoform analysis after quantification with Salmon/Star

Question

[rajdeepboral00]

Hello all, I have been given with STAR aligned BAM files, so I can use STAR/SALMON to get the transcript level counts. Basically I need to focus on the isoform level changes , so what are some good tools to explore that? And which one will be better? Counting with STAR or SALMON ?

Answer 1

[i.sudbery]

It depends on what level you would like your analysis to take place on. In general, if you are interested in complete isoforms, then Salmon gives you isoform level quantification, where as STAR doesn't. However many tools that operate on alternate splciing level, require either exon or junction counts, or BAM files as input, which would mean using STAR.

Decide what you are intersted in:

• Differential Transcript Expression (DTE): This simply compares the level of each transcript in each sample. So if a gene has two isoforms iso1 and iso2, and they are measured in two conditions A and B, DTE would give you two numbers: LFC[iso1] = count[iso1,condA]/count[iso1,condB] and LFC[iso2] = count[iso2,condA]/count[iso2,condB]. That is iso1 and iso2 are treated as if they are entirely independent of each other, and their fold changes measured seperately. Any resulting log fold change could be due to changes in transcription, or changes in splicing or transcript stability. The most common tools here are edgeR/limma/DESeq2, and you would want Salmon quantification.

• Differential Transcript Usage (DTU): This compares the fraction of a gene that is an isoform in one condition to the fraction of a gene that is that isoform in the other gene. In the example abouve it is DTU[gene] = (count[iso1,condA]/count[iso2,condA])/(count[iso1, condB]/count[iso2, condB]). Any log fold change here could be due to a switch from one isoform to another, (i.e. iso1 down, iso2 up), such as a change in splicing, or it could be due to a differnce in the level of only one of the isoforms (e.g. iso2 is destablised, but iso1 is unaffected). It may or may not be accompanied by a overall change in the level of the gene, but that will not show up in the analysis. The most common tools here are IsoformSwitchAnalyzeR/DEXSeq(on transcript counts)/limma::diffSplice. Again, you want Salmon quantification. You would also want to generate your Salmon quantification with pseduoreplication.

• Differential Exon/junction Usage (DEU): Here the inclusion ratio of each indevidual exon in each of isoform 1 and 2 is compared between the conditions. Again, this won't distinguish between a switch from one form to the other and just one form changing levels but not the other. Although I've reffered to it as DEU, the same principle applies to junctions. The upside here is that you get results on indvidual events, like exon inclusion or skipping. There is also much less usage of inferential statistics on isoform levels from reads that could come from multiple transcripts - it just uses straight forward counts, so in that way the results are more reliable. They can however be hard to interpret. If an exon is 10 times more included in condition A than condition B, that might seem exciting, but not when it turns out that both the inclusion and exclusion forms only make up 10% of all transcripts from a gene. Top tools here are rMATS (uses BAM files as input), SUPPA2, MISO, DEXSeq (on exon or junction counts).

Which is best probably depends on what you want out of your analysis.

[rajdeepboral00]

Thank you for your detailed reply

Answer 2

[Gordon Smyth]

The following papers show the advantages of Salmon with Gibbs resampling, followed by edgeR's catchSalmon, glmQLFit, glmQLFTest (for DTE) and diffSplice (for DTU):

Baldoni PL#, Chen L#, Li M, Chen Y, Smyth GK (2025). Dividing out quantification uncertainty enables assessment of differential transcript usage with diffSplice. bioRxiv https://doi.org/10.1101/2025.04.07.647659

Baldoni PL, Chen L, Smyth GK (2024). Faster and more accurate assessment of differential transcript expression with Gibbs sampling and edgeR v4. NAR Genomics and Bioinformatics 6(4), lqae151.

Baldoni PL#, Chen Y#, Hediyeh-zadeh S, Liao Y, Dong X, Ritchie ME, Shi W, Smyth GK (2024). Dividing out quantification uncertainty allows efficient assessment of differential transcript expression with edgeR. Nucleic Acids Research 52(3), e13. https://doi.org/10.1093/nar/gkad1167