challenges in implementing modularityFrom virtual genomes to virtual cells The genome projects have given us a virtual genome: molecular biologists have been able to abstract a biological object, the DNA of an organism, into a mathematical object, a "word" in an alphabet of four letters. It is no longer necessary to do wet lab experiments on the biological object; a computational infrastructure makes it possible to sit at a computer and analyse a virtual genome.
What comes next? With the development of systems biology, we want to construct mathematical abstractions, not of the structure of a single molecule, but of the behaviour of collections of interacting molecules. This is a much harder challenge.
A vision of incremental, modular model building We would like to construct such mathematical models in an incremental and modular fashion. A biologist interested in the EGF signalling pathway should be able to construct a model of it using parts already constructed by others—one part describing the receptor and the intricacies of receptor dimerisation and endocytosis, a second part describing the RAS GTPase switch, a third describing the MAP kinase cascade and its scaffold proteins, a fourth describing Elk1 and its activation of target genes—along with more specific assumptions describing the hypotheses which the biologist is exploring (perhaps the existence of a hitherto unsuspected feedback mechanism), and then dropping these parts and assumptions into a "virtual cell", much as a biochemist might drop the actual ingredients into a test tube. All this at the level of biological, rather than mathematical, description. The computational infrastructure should then wire the parts together to produce a working mathematical model of the pathway†.
†"Perhaps one day, building a model such as the one presented by Klipp et al will be as simple as downloading a phospho-relay module, a model of the HOG pathways and glycolysis, linking them together with some generic gene expression circuitry and a custom turgor pressure formula, and presto ... Systems Biology"
Comment by Patrick d'Haeseleer in "Closing the circle of osmoregulation", Nature Biotechnology 23:941-2 2005.
Edda Klipp et al, "Integrative model of the response of yeast to osmotic shock", Nature Biotechnology 23:975-82 2005. PubMed
This is not how models are currently built. Models are usually constructed by those adept at wrestling with differential equations, who mostly build them as monolithic entities. They may be wrapped in SBML‡ which enables different tools to access the same model. SBML has played an important role in nucleating a modelling community and has become a de facto standard. While it enables reuse, the choice of a data exchange format based on XML, which treats models as data types, limits its capacity for modularity. With or without SBML, it remains hard to build models incrementally and harder still to share parts of one model with another in a modular way. Indeed, it is hard enough simply to understand a model built by someone else, as one quickly discovers when digging into the Supplementary Information in which such models are always buried. (This is so much the case that one cannot help wondering if most reviewers have ever attempted it!) Until models can be used in different contexts (as modules in more complex models), they cannot become credible scientific objects upon which the rest of us can rely. This perhaps explains why most biologists have not yet incorporated models into their working practice.
‡Mike Hucka et al, "The systems biology markup language (SBML): a medium for representation and exchange of biochemical network models", Bioinformatics 19:524-31 2003. PubMed
One problem with models of molecular systems is that they can get very complicated very quickly. Of course, it is much better if we can simplify these into something manageable, as we do in physics. Unfortunately, we do not yet understand how to do this systematically in biology. Perhaps we will, eventually, but for the moment we either stick to simplifiable systems or we confront biological complexity and learn to live with it. We take the latter approach here.
challenges in implementing modularity