Addressing the challenges of multiscale model management in systems biology

Abstract

Mathematical and computational modelling are emerging as important techniques for studying the behaviour of complex biological systems. We argue that two advances are necessary to properly leverage these techniques: firstly, the ability to integrate models developed and executed on separate tools, without the need for substantial translation and secondly, a comprehensive system for storing and man-ageing not only the models themselves but also the parameters and tools used to execute those models and the results they produce. A framework for modelling with these features is described here. We have developed of a suite of XML-based services used for the storing and analysis of models, model parameters and results, and tools for model integration. We present these here, and evaluate their effectiveness using a worked example based on part of the hepatocyte glycogenolysis system.

Introduction

Modelling physiology is in many ways similar to the modelling of process systems so there is much that chemical engineers can contribute. As with process systems, one of the major challenges in computational physiology is to efficiently integrate existing computational models which describe phenomena associated with a variety of spatial and temporal scales. Such models can be deterministic, stochastic, qualitative, or in many other forms. An important part of this challenge is the storage, collation, and retrieval of models, along with their integration.

Our work (The UCL Beacon Project, 2002–2007) is part of the UK Department of Trade and Industry sponsored Beacon program, focused on harnessing genomics. We aim to build in-silico models that represent aspects of behaviour of the human liver, an epithelial organ. The methodology and modelling system should then be extendable to other epithelial organs. In building a fully integrated model of the liver, existing models of various components must be used along with newly devised models. Our approach is therefore to develop a system for the orchestration and integration of models. Not only will this system permit the development of integrated models which could not otherwise be constructed, it will also support the development of these models in a manner which increases the computational efficiency and reliability of those models, and reduces the time taken for such development.

The framework we have developed supports two key aspects of biological modelling: model integration across different scales, and the interconnection of the distinct components in biological systems. Interconnections are largely based on signalling i.e. the transport and reaction of chemicals between distinct components that drive the physiological system. Using this framework we aim in the project to develop a simulation environment in which a wide variety of models are integrated and exploited within a common domain of interest. These models may be at different levels of abstraction, may deploy different representations, and may focus on different interacting phenomena. Validation may give rise to model variants that will require management.

Our project will result in a system to integrate models addressing phenomena from the level of individual gene and cell features through tissue and organ models. Models at every level of the structure will be integrated, validated, and exploited using a plethora of mathematical, computational and experimental techniques. Fig. 1 shows the hierarchy of levels of signalling activity in many physiological systems.

One of the fundamental issues in model integration is how to handle the intrinsic inter-relationships between different models in an efficient way. Individual models are built up in an isolated biological environment relative to the real physiology and the purpose of linking different models is to recover the physiological conditions in terms of the context the models cover. Our computational framework for linking biological models will take account of the intrinsic couplings existing among the models, while allowing the flexibility that comes from being able to ‘plug’ in different choices of model, and link models which take different approaches to modelling, or which apply to different scales of consideration.

In this paper, we shall review existing work on computational infrastructure for systems biology, argue that two areas of software engineering (information management and encapsulation) should in particular be brought to bear upon the problem and describe a series of software modules we have authored that together constitute a complete computational environment for systems biology. In particular, the system supports the integration of models built in very different software environments while leaving the authoring and execution of the component models within those environments. We provide evidence for the effectiveness of our technique using an example model of part of the response of the liver to adrenaline, where one of the component models is built in Mathematica, and another in X-Phase-Plane-Auto (XPPAUT).