Bayesian modelling of gene expression from high-throughput sequencing experiments (RNA-seq)

Abstract

Gene expression measurements are widely used in biology to study the function of proteins in living organisms. Differential expression between for example different tissue types, or between healthy and diseased individuals, can help understand which proteins and molecular pathways are involved in different biological processes.

Gene expression is often measured using high-throughput sequencing techniques (known as RNA-seq) which result in millions of measurements of short sequences of RNA molecules. These data are highly complex and require a significant amount of processing before measurements can be used in statistical analysis comparing expression between different individuals or experimental conditions.

We present a Bayesian model for gene expression from RNA-seq experiments. Our model enables estimation of expression of different isoforms (different versions of the same gene) and also haplotype-specific isoforms (the two different inherited parental versions of the gene for an individual). We apply the method to data from inbred mice, in which we can detect imprinted genes, for which only the version from one parent is expressed.