Bioinformatics Seminar
Time: 11AM
Venue: Davis Auditorium and Online
22 September 2026
This is a WEHI only event.Transcripts and other stories: realistic RNA-seq simulation for transcript-level differential expression
Pedro BaldoniWEHI
Differential expression analysis of RNA-seq data is increasingly carried out at the transcript rather than the gene level, where changes in isoform usage can be detected directly. In this talk, I will present recent developments in edgeR and limma for transcript-level analyses, how quantification uncertainty is carried into differential testing, and a realistic RNA-seq simulation framework. Methods development relies on ground truth, and simulations are often its only practical source. Realistic simulation has become more demanding as transcriptome annotations grow more complex. Using real human and mouse experiments, I will characterise the data generating mechanism of transcript expression, in particular the biological coefficient of variation and its trend with abundance. I will then describe a probabilistic model for PCR duplication at the sequencing read level. Treating amplification as a branching process and sequencing as random sampling yields the exact distribution of duplicate copies of observed fragments. This lets us assess how DGE, DTE and DTU methods perform when PCR duplication is present. Finally, I will show the extent to which strand-specific sequencing reduces read-to-transcript ambiguity for transcripts that overlap on opposite strands. These results argue for building and evaluating transcript-level methods against simulations that reproduce sequencing noise and transcriptome complexity.