 |
Center for Nonlinear Studies |
RNA and protein expression are discrete and stochastic processes, where randomness in the motion or activation of single molecules can determine the fate of a cell or organism. As more and more techniques are developed to experimentally perturb and measure single molecules in individual cells, new computational tools must be developed to rigorously integrate these highly informative data with models describing the underlying biochemical dynamics. This work describes two new computational methods that help to close the loop between computational models of stochastic gene expression and single-cell experiments. The first method provides new bounds on the likelihood of single-cell gene expression data given a stochastic model formulated with the chemical master equation. These bounds use single-cell measurements to determine the model precision necessary to distinguish between more or less likely model parameters. We demonstrate this tool by identifying a predictive model of transcription activation in S. Cerevisae from single-molecule fluorescent in-situ hybridization (smFISH) experiments.
The second method uses predictive models of stochastic gene expression to design more informative single-cell experiments. The Fisher information matrix (FIM) has been used in a variety of other engineering applications to design efficient experiments and improve estimates of parameters. Existing Fisher information methods assume normally distributed noise statistics or sufficiently large data to invoke the central limit theorem. However, for most single-cell analyses, distributions of biomolecules are non-Gaussian or data sets are limited in size, situations that directly violate these assumptions. We demonstrate this experiment design process to the same model of transcription activation in S. Cerevisae. By systematically designing experiments to use all the fluctuation information within modern single-cell data, our methods can improve co-design of experiments and models to enhance the efficiency and accuracy of quantitative systems biology.
Host: William Hlavacek