DEXSeq for isoform data

Question

[inah]

Hi, I have a question about the application of DEXSeq to isoform data (estimated counts). EdgeR and DESeq2 were designed for gene counts, not isoform counts, and they use either a common dispersion parameter for all genes, or a common prior (essentially). EBseq was designed for isoform data and allows the dispersion parameter prior to differ among different groups of isoforms (groups formed based on level of uncertainty). DEXSeq appears to use a different dispersion parameter for each isoform (counting bin within gene), but does not use different priors as in EBseq. Is there an issue here? Thanks, Ina

[ATpoint]

edgeR can handle isoform data if starting from salmon or kallisto quantifications with inferential replicates. The catchSalmon or catchKallisto can use the bootstrap information to calculate a per-transcript overdispersion value to take the mapping uncertainty into account. With this you can then perform the standard DE pipeline.

[i.sudbery]

The difference with DEXSeq is that it compares different isoforms of the same genes, and then asks if the ratios of isoforms changes between conditions, rather than comparing one isoform between two conditions. Thus EBseq and DEXseq are doing different things.