Open Collections

Energy Lasso: Combining Rule-Based Modeling and Bayesian Parameter Estimation to Infer Biochemical Mechanisms

Banff International Research Station for Mathematical Innovation and Discovery

Rule-based modeling (RBM) is a modular and scalable approach to specifying and simulating large- scale models of cell regulatory networks. How can such models be related to experimental data and, more importantly, used to make mechanistic inferences?\\r\\n\\r\\nWe start by presenting a generalized form of the thermodynamic rule-based modeling called energy BioNetGen Language (eBNGL), which builds on work of Ollivier and Swain and of Danos and colleagues. The key idea is that cooperative effects between sites of transformation (e.g., binding and post-translational modification) can be encoded as energy parameters, which, when non-zero, introduce variation in the rates of reaction over the initial reaction classes. In RBM each rule serves as a generator of a specific class of chemical transformation, e.g., a non-covalent bond between two sites of different proteins or a posttranslational modification of a particular site, and all reactions generated by a rule inherit the same rate constant. eBNGL allows this assumption to be relaxed and permits systematic investigation of cooperativity parameters. Although these parameters cannot typically be measured, pioneering work by Sethna and colleagues has shown that biochemical models typically exhibit a characteristic known as “sloppiness,” which is a wide dynamic range in the sensitivity of a model’s behavior to its input parameters. This property has the counterintuitive effect that even models that have many parameters that are poorly constrained by data can yield relatively tight behavioral predictions. Furthermore, mechanistic inferences may be made on the basis of the subset of parameters that are tightly constrained. Work by a number of groups has shown that Bayesian parameter estimation techniques perform well in parameterizing such complex models, carrying out model selection, and inferring mechanisms.\\r\\n\\r\\nWe have coupled eBNGL and Bayesian parameter estimation with the goal of identifying cooperative mechanisms in signal transduction networks. Model selection is performed by introducing L1-regularization – aka LASSO – which holds cooperativity parameters to a zero value unless evidence suggests otherwise. We apply this approach to Kholodenko’s data on phosphorylation of the epidermal growth factor receptor (EGFR) in hepatocytes, which was previously used to infer a strong negative cooperativity between phosphorylation of the adaptor protein Shc and its binding to phosphorylated EGFR. We find that positive cooperativity between the binding of Shc and another adaptor protein, Grb2, to phospho-EGFR can also account for the observed kinetics, and we are currently applying model selection procedures to determine whether one of these mechanisms is favored by the data.

Mathematics; Biology and other natural sciences; Systems theory; control; Mathematical biology

video/mp4

Author affiliation: University of Pittsburgh

2014-08-07

Attribution-NonCommercial-NoDerivs 2.5 Canada

http://hdl.handle.net/2429/49621

Unreviewed

http://creativecommons.org/licenses/by-nc-nd/2.5/ca/