Accelerated Bayesian inference from snapshots of single-cell transcripts

This paper is an improvement upon BayFish, our software for Bayesian inference of transcription dynamics from population snapshots of singlemolecule RNA FISH in single cells. However, Bayesian inference is computationally expensive and cannot tackle more complex models of stochastic gene expression in a reasonable amount of time.

To address this problem, we developed a Poisson Mixture Piecewise Deterministic Markov Switch Rate (PM-PDMSR) algorithm that is 1000-fold more efficient than the state-of-the-art algorithms commonly used to solve the time-dependent dynamics of complex models of stochastic gene expression. With this efficiency, we performed a fullscale Bayesian analysis including the parameter identification, uncertainty quantification, and most importantly, an estimation of the Bayesian evidence for a given gene expression model (compared to other competing complex models).  This was a fruitful collaboration with Yen Ting Lin at Los Alamos National Laboratory.