Next: Further reading Up: A diversity of models Previous: Logical, continuous and stochastic
Subsections
Different motivations for building models
Finally, models may be differentiated by the question that motivates their development. This motivation may range from a desire to obtain quantifiable values for some aspect of a system that can then be experimentally validated through to exploring high-level principles of cellular control. While it can be an oversimplification to categorise a model according to its purpose, as many models will overlap across different categories, several broad approaches can be discerned.
Crafted models of specific systems
Many attempts at modelling regulatory interactions focus on small, well understood systems that can be modelled by hand from available empirical knowledge. These models generally have a high level of fidelity to the underlying biological system, with each component in the model system corresponding to a particular element of the biological system. Numerical and computer simulations are used to make predictions about systems that are too complex to allow for analytical solution. This category includes both continuous models, such as the various models of phage-
[109,102,9], as well as logical models, such as Bodnar's Boolean characterisation of Drosophila embryogenesis [18].
Phenomenological models of biological mechanisms
Another approach, at a slightly higher level of abstraction, is to use systems of generalised components to reproduce observed biological behaviour, such as morphogenesis and pattern formation. In these models, there is no longer a direct mapping between components in the model and copmonents in the biological system, however the high level behaviour of the system is preserved. An example of this category is the gene circuit models developed by Mjolsness, Reinitz and Sharp for modelling segmentation in Drosophila [87].
General models of classes of networks
Other researchers, rather than investigating individual systems, have taken the approach of characterising the behaviour of classes of networks with particular structural and dynamic properties. These approaches frequently work with simplified descriptions of gene activation that allow much larger and more complex networks to be simulated than would otherwise be possible. A common technique is to generate a large number of random networks (an ensemble) governed by a specified set of local rules and observe the statistical properties of the global behaviour [69]. Another type of modelling that falls into this category is the exploration of networks whose structures share particular statistical properties, such as scale-free connectivity distribution [12], or hierarchical patterns of modularity [98].
Network models inferred from experimental data
The rapid increase in available experimental data in recent years has shifted some of the focus towards techniques that are able to automatically construct models of larger, less well-understood regulatory systems. Advances here are divided between both the formalism that is used to model the system, and the learning algorithms that are used to derive the model from the available data [134].
Next: Further reading Up: A diversity of models Previous: Logical, continuous and stochastic Nic Geard 2004-05-06