A hierarchical negative-binomial model for analysis of correlated sequencing data: practical implementations.

Abstract

High-throughput techniques for biological and (bio)medical sciences often result in read counts used in downstream analysis. Nowadays, complex experimental designs in combination with these high-throughput methods are regularly applied and lead to correlated count-data measured from matched samples or taken from the same subject under multiple treatment conditions. Additionally, as is common with biological data, the variance is often larger than the mean, leading to over dispersed count data. Hierarchical models have been proposed to analyze over dispersed, correlated data from paired, longitudinal, or clustered experiments. We consider a hierarchical negative-binomial model with normally distributed random effects to account for the within- and between-sample correlation. We focus on different software implementations to allow the use of the model in practice.