Developing variational Bayesian inference for applications to gene expression data
David Walker · 2021
Bayesian hierarchical generalized linear models are intuitively appealing for applications to gene sequencing data. However, they can be computationally costly to fit in high-dimensional settings using standard Markov-chain Monte Carlo methods. Here we explore the use of variational inference techniques to approximate the posterior of Bayesian hierarchical GLMMs for detecting differential expression in RNA-Seq data or differential translational efficiency in Ribo-Seq data. We find that in simulation studies the variational approach is comparable to two common methods for detecting differential expression, and that the variational posterior is close to the Markov-chain Monte Carlo posterior.