Incorporating extrinsic noise into mechanistic modelling of single-cell transcriptomics

A mechanistic understanding of single-cell transcriptomics data requires differentiating between intrinsic, extrinsic and technical noise, but an abundance of the latter often obscures underlying biological patterns. Accurately modelling such data in the presence of large cell-to-cell heterogeneity due to factors such as cell size and cell cycle stage is a challenging task. We propose a tractable, fully Bayesian framework for mechanistic modelling of single-cell RNA sequencing data in the presence of cellular heterogeneity. Applied to murine transcriptomics data, we show that cell-specific effects can significantly alter previously inferred dynamics of individual genes. Our implementation is statistically exact and readily extensible, and we demonstrate how it can be combined with Bayesian model selection to compare various models of gene expression and measurement noise.