We introduce model-based Bayesian inference to screen for differentially expressed genes based on RNA-seq data. RNA-seq is a high-throughput next-generation sequencing application that can be used to measure the expression of messenger RNA. We propose a Bayesian hierarchical model to implement coherent, fast and robust inference, focusing on differential gene expression experiments, i.e., experiments carried out to learn about differences in gene expression under two biologic conditions. The proposed model exploits available position-specific read counts, minimizing required data pre-processing and making maximum use of available information. Moreover, it includes mechanisms to automatically discount outliers at the level of positions within genes. The method combines gene-level information across replicates, and reports coherent posterior probabilities of differential expression at the gene level. An implementation as a public domain R package is available.
Keywords: Bayes; Differential Gene Expression; FDR; Mixture Models; Next-Generation Sequencing.