Bayesian inference of differentially expressed transcripts and their abundance from multi-condition RNA-seq data

Deep sequencing of bulk RNA enables the differential expression analysis at transcript level. We develop a Bayesian approach to directly identify differentially expressed transcripts from RNA-seq data, which features a novel joint model of the sample variability and the differential state of individual transcripts. For each transcript, to minimize the inaccuracy of differential state caused by transcription abundance estimation, we estimate its expression abundance together with the differential state iteratively and enable the differential analysis of weakly expressed transcripts. Simulation analysis demonstrates that the proposed approach has a superior performance over conventional methods (estimating transcription expression first and then identifying differential state), particularly for lowly expressed transcripts. We further apply the proposed approach to a breast cancer RNA-seq data of patients treated by tamoxifen and identified a set of differentially expressed transcripts, providing insights into key signaling pathways associated with breast cancer recurrence.