Question about funnel-shaped log2FC–dIF pattern in DEXSeq

[hannalee809]

Hi all,

I have a quick question about the expected distribution from IsoformSwitchAnalyzeR + DEXSeq. When plotting gene log2FC vs dIF after isoformSwitchTestDEXSeq(), I usually see a symmetric distribution centered around log2FC = 0 and dIF = 0 for most biological contrasts. However, in one near-control comparison (scrambled vs naïve), the distribution is asymmetric: gene expression appears globally shifted toward one condition, and the points form a funnel/triangle shape with increased dIF spread and more significant switches than expected.

Is this pattern typically indicative of technical effects (e.g., library size/composition differences, low-expression isoforms, or insufficient filtering) rather than true switching? Are there recommended filtering or normalization steps to avoid this behavior in near-null comparisons?

Thanks for any insight!

[elena-iri]

Hi @hannalee809 !
Thank you for using our package and sharing your doubt!
This does indeed look strange... Not everything should appear down-regulated in the differential gene expression analysis. Perhaps you should double-check how you obtained the gene log2 fold.
Best,
Elena