Estimation of isoform expression using hierarchical Bayesian model by RNA-seq

Abnormal expression of isoforms may cause complex human diseases. So it is crucial to estimate isoform expression and identify the abnormal isoforms. RNA-seq technology enables us to monitor the expression on genome-wide scale at single base pair resolution. Many methods are developed to estimate isoform expression based on this technology. But all these methods treat gene expression as a by-product by the way of summing its corresponding isoforms expression and don´t exploit the relationship between gene and its isoforms expression. In this paper, we develop a new hierarchical Bayesian method to estimate isoforms expression. In this model, we treat the gene expression as the sum of its corresponding isoforms expression, and we explicitly model the gene expression level and the relationship between gene and corresponding isoforms using Multinomial distribution. By this way, gene expression is treated as a variable and can be included in a unify framework. More importantly, the relationship between gene and its isoforms help us to achieve a higher accuracy on isoform expression level. The simulation and real data studies show that our method is more effective than other state-of-the-art algorithms for isoform expression estimation.