Bioinformatics Seminar
Time: 11AM
Venue: Davis Auditorium and Online
3 November 2026
This is a WEHI only event.What is Gene Expression?
Hasini HerathWEHI
Gene-level differential expression analysis of RNA-seq data is increasingly based on transcript abundances estimated by tools such as Salmon, kallisto and RSEM, rather than on traditional read counts. Short reads are often compatible with multiple transcripts, so these transcript-level estimates carry uncertainty arising from read-to-transcript ambiguity (RTA). Gene-level differential expression analyses are generally considered more stable than transcript-level analyses, because aggregating transcript estimates to the gene level removes most of this uncertainty. However, some uncertainty remains when reads are compatible with transcripts from different genes. This research investigates how quantification uncertainty and the completeness of the reference annotation affect gene-level differential expression analysis within the edgeR or limma frameworks. When genes express multiple transcripts, the summed counts exhibit considerable technical overdispersion, which inflates the dispersion estimates obtained in edgeR. Dividing gene counts by the gene-level RTA overdispersion, estimated from inferential replicates, reduces this inflation and increases statistical power while maintaining control of the false discovery rate. A comparable reduction in dispersion was observed in the experimental datasets. Future work will determine whether effective-length offsets and RTA overdispersions require sample-specific estimation, including empirical Bayes moderation of the offsets, and will evaluate the effects of incomplete and expanding annotations on quantification and downstream inference.