#C0493FQQZ3P I've noticed some discrepancies in transcript expression results between DEXseq and other tools (Salmon/IsoformSwitchAnalyzeR). E.g. for ENSMUST00000196245.1 DEXseq: KO4 (3.02) > WT (0.01)

groupID    featureID    exonBaseMean    dispersion    stat    pvalue    padj    KO4    WT    log2fold_WT_KO4
ENSMUSG00000005057.13    ENSMUST00000196245.1    190.507021901714    0.10224272045332    89.1840555286677    3.59736677761298e-21    4.94427051800339e-19    3.02205908654677    0.0142198797972508    -15.4717752607727

IsoformSwitchAnalyzeR: KO4 (0.37) < WT (9.79)

"","iso_ref","gene_ref","isoform_id","gene_id","condition_1","condition_2","gene_name","gene_biotype","iso_biotype","gene_overall_mean","gene_value_1","gene_value_2","gene_stderr_1","gene_stderr_2","gene_log2_fold_change","gene_q_value","iso_overall_mean","iso_value_1","iso_value_2","iso_stderr_1","iso_stderr_2","iso_log2_fold_change","iso_q_value","IF_overall","IF1","IF2","dIF","isoform_switch_q_value","gene_switch_q_value","PTC","IR","switchConsequencesGene"
"3801","isoComp_00063012","geneComp_00020098","ENSMUST00000196245.1","Sh2b2","KO4","WT","Sh2b2","protein_coding","protein_coding",12.2176627648093,7.35997484868363,24.9504302971442,0.718745975428075,2.1933373408698,1.75991120428755,NA,3.39185008894742,0.378354585004484,9.79719568183778,0.378354585004484,2.04034647014029,4.65839426007501,NA,0.142566666666667,0.0431666666666667,0.384533333333333,0.341366666666667,0.0215286797891844,0.00165001636314129,FALSE,0,TRUE

Salmon showed similar counts to IsoformSwitchAnalyzeR. What could cause this problem? Which analysis is more credible? Thanks for any help.

Are you sure you know if

*_value1

is WT or KO4 in IsoformSwitchAnalyzeR? I think some tools use different conventions from other tools, so things might just be switched. And each tool has different criteria for including/excluding reads and measures something slightly differently, so it is not surprising the values are different. If you really want to know what is going on with a specific event, I know of no substitute for comparing the BAM files directly, e.g. by looking at read coverage and splice junctions in a viewer like IGV. This will give you an idea of how "credible" an event detected by a tool is, and if the data really supports that particular event, or really something else that the tool is misinterpreting. (As an example, I had events that DEXSeq interpreted as differential inclusion of an exon, but when looking at the site in IGV, it turned out that the entire intron was retained, not just the supposed exon. And you are looking at entire transcripts, but there may well be differential events that tools interpret as supporting some annotated transcript, but in reality they might be from some transcript that is not annotated at all.)

Hi @Thomas Danhorn! I checked several other genes and transcripts in IGV, and it appears that the results from IsoformAnalyzeR and Salmon are more consistent with read coverage in IGV, while DEXseq results often contradict this. However, the difference in exon expression calculated by DEXseq seems still valid. Therefore, I guess the issue comes from using DEXseq to calculate transcript differential expression.

I mostly use DEXSeq to detect differential exons rather than full transcripts, but I think it generally subtracts out any gene expression effects and just compares transcript ratios. Whereas Salmon gives you absolute values for each transcript. Example: Group A (e.g. control): Transcript1: 12 Transcript2: 6 Group B (e.g. treatment or mutant): Transcript1: 3 Transcript2: 6 DEXSeq compares ratios (e.g Transcript2 has 33% in group A and 67% in group B), whereas Salmon would rate Transcript2 as "unchanged", Transcript1 as 4x down, and overall gene expression as 2x down. So you have to make sure you double-check and understand what each tool is actually measuring before interpreting the results. Especially if the overall gene expression changes together with the transcript ratios, this can be quite confusing.