TR&D3 Dynamical Modeling

University Of Connecticut Sch Of Med/Dnt, Farmington CT

Investigators

Linked publications & trials

Abstract

TR&D3 Dynamical Modeling Project Summary Experimental data now accumulates at a rate that far exceeds the current ability to assimilate that data, and the number of elements and complexity of mathematical models is rapidly increasing to understand and explore that data within the context of complex biological mechanisms. However, large mechanistic models are difficult to build, understand and validate, often resulting in poor predictive value of the models. The challenges with highly complex models are: (1) the structure of large network models can be difficult to understand, (2) large models are almost always under-constrained, having not enough data to measure or fit numerical parameters, and (3) the underlying biological mechanism are either not well established or differ in different contexts. These challenges can severely limit the predictive power and usefulness of the models and simulations. We propose two aims two address these challenges. To help build, organize, and navigate complex models, we propose to create ModelBricks, fully annotated modules that can be added to a model, and assembled into larger models. The introduction of a rule-based modeling approach into VCell, coupled with the integration of VCell with public databases via Pathway Commons, provides the basis for developing a new modular model building capability that will leverage both the power of the VCell database and community resources. We will develop a VCell architecture for ModelBricks that creates the fundamental design elements required to enable packaging of model elements into structures that can be queried in the database and used within VCell BioModels or exported to other modeling formats. An impressive set of eleven DBPs will help the VCell team to seed a growing library of common ModelBricks of common physiological processes such as ion pumps and channels, phosphorylation reactions, and GTP/GDP cycling of G proteins. This initial library will serve as a core for community-driven development: ModelBricks will be a powerful aid to building models of all types and complexity, and spawn the development of well-annotated models that can be searched, reutilized and recombined. A second aims attacks the problem that the traditional approach of fitting parameters to experiment fails when limited amounts of experimental data do not sufficiently constrain the parameter space of large models. There exist, however, novel sophisticated analytical techniques that can help validate and discover unexpected predictive powers of such models. Three approaches will be developed within VCell to address model uncertainty; (1) methods for steady-state analysis such as steady state computation, stability analysis, sensitivity analysis; (2) ?Sloppy Model? approaches that derive parameter combinations that drive model behavior in the context of specific observables of interest even when unconstrained; and (3) methods to explore uncertainty in model structure using simulation and analysis of ensembles of closely related model variants matched to specific observations. These approaches will rely on strong technical collaborations with Mendes and Brown and be driven by DBPs with Bader, Fournier, Heinen, Mayer, and Iyengar.