Biophysically informed modeling for mapping the effects of genetic and environmental perturbation on cell states

Improvements in single-cell genomics and scalable genetic manipulation have enabled the development of high-throughput, combinatorial perturbation experiments at single-cell resolution. Such advances present an opportunity to understand the impact of perturbations on the biophysics of transcription and resulting cellular states. However, current predictive models are largely phenomenological, limited to treating average expression levels, rather than causal mechanisms. Principled mechanistic analysis requires discrete, stochastic models of biological and technical noise.

We develop a mechanistic, chemical master equation-based modeling framework for predicting perturbation responses in unseen conditions from single-cell datasets. Our approach fits biophysically meaningful models of transcription using splicing data available in these genomics experiments, parametrizing the cellular response by three kinetic rates as opposed to mean expression. We predict perturbation responses as changes in these parameters, which reveal the differential impacts of perturbations on molecular production and processing. This approach enables targeted, hypothesis-driven research on the causal properties of perturbations on cellular phenotypes, guided by mechanistically-grounded modeling.