Statistical Characterization of Biochemical Network Models
- Author(s): Biddle-Snead, Charles
- Advisor(s): El-Samad, Hana
- et al.
Complex systems of numerous interacting biomolecules dictate cellular behavior. To better understand how these systems operate in aggregate, computational models of these systems may be constructed, aligned with experimental data and analyzed to reveal high-order system functionality and produce hypotheses for further experimentation. Different model frameworks offer trade-offs in terms of scale and detail; here, to capture quantitative details of single-cell gene expression data, and to gain insight into the systems generating that data, we use dynamic biochemical reaction networks (BRN) represented by systems ordinary differential equations (ODE). At this scale, challenges arise from both the ODE model framework and the detailed properties of the modeled systems. On the model side, difficulty stems from lack of knowledge of model parameters, their relative certainty when estimated, and from disentangling the often complex relationship between model structure and function. On the system side, behavior is impacted from the small scale details of biomolecular physics as well as the larger scale cellular context in which a system of interest operates.
Here, we develop a computational framework to infer data and function constrained parameter posterior distributions and show that these distributions can be used to address challenges in both model analysis and complex system behavior. We apply these methods to models of three biochemical systems: chiefly, we studied the regulatory network controlling flexible GAL1 gene expression in yeast. Our computational approach, together with detailed experimentation, reveals the systems level basis for heterogenous, context-specifc decision making in this circuit.
The proposed framework has implications for understanding and quantifying context dependence in natural biochemical networks as it arrises in cell type differentiation, tumorogenesis and elsewhere, and rational control of natural systems through multi-component therapies. Additionally, our approach is a potentially powerful one in the rational design and control of synthetic systems.