Systemic computation is designed to be a modelling language of natural systems. Using SC a biological system can be analysed and a hypothetical mechanism for its functioning can be developed as a formal SC model. The same mechanism can then be run as a simulation, its simulation behaviour providing the predicted behaviour for the hypothetical mechanism. With this approach, all that remains is theorizing (is the model right?) and comparing (does the model behave correctly?

Modelling and simulation is still something of a black art. While many researchers are trained in experimental design and statistical analysis, the majority of biology modellers are self-taught because of the interdisciplinary nature of the work. The result can be a mixture of mathematical and software models, not fully understood by biologists and potentially not incorporating the very concepts they are supposed to help analyse. While not every researcher may be comfortable with using SC for their application, lessons learned from the use of these techniques can be exploited by all those with an interest in modelling:

  1. Identify the scope of your work. Explicitly identify which biological system is to be the subject of study and which behaviour is of relevance. This is often best performed by examining the availability of data — the more reliable the data that are available, the better the model will be. This should not just be data describing the biological behaviour; it should be as much data as possible on the nature and workings of the biological system itself, its constituents and interactions between them.
  2. Explicitly state your hypothetical mechanism: how do you believe the biological system is working; what assumptions are you making; what predictions can you make about the behaviour if your hypothesis is correct; and are there any behaviours that, if observed, would disprove your hypothesis? Again, this should be stated with as much detail as you can.
  3. Design the simulation (which when using SC you can achieve by coding your hypothetical mechanism above). A simple design is often easier and desirable, but not always better — biology is not simple and so the simplest models may lack significant detail. Ensure that you do the following (which in SC is called the Systemic Analysis):
    1. Define the level of abstraction and try to represent everything at that level. If you are trying to model tumour development, you may want to model cells and some genes and proteins — you do not need to model electron clouds around atoms, or populations of evolving organisms. Or if you are modelling an ecology, you may model populations of organisms and weather patterns, but you can ignore cells and proteins. However, do not forget that all of these features exist in reality and may form the context that modifies the interaction of entities at the level of abstraction you have chosen. The decision of which level of abstraction to use is often determined by the amount of data available and the feasibility of computation.
    2. Identify the features to be modelled. In SC, this equates to identifying the systems. So if modelling tumours: which cells will be included in the model, which genes, and which proteins? These may include some very fundamental features such as the laws of physics, Brownian motion or reflectance of light. The normal rule of thumb is to include those systems that have the most effect. But everything not explicitly modelled may still have an effect, so try to state and justify what is not in the model in addition to what is in it.
    3. Identify the organization of the features. In SC, this corresponds to specifying the scopes of systems: which systems are able to affect others and which are not; which contain subsystems or may subsume or expel systems in the future; and how are systems physically arranged in relation to each other?
    4. Identify the interaction of the features. In SC, this corresponds to specifying the interacting systems and the contexts for those interactions. In real life, nothing happens without some interaction. The most revealing models are those that produce behaviours because of the interaction of their constituents, for such models provide better explanations for the cause of those behaviours. (For example, a numerical model that describes evolutionary dynamics with a single equation may be very useful, but a model that describes evolutionary dynamics through modelling of interacting individuals allows the analysis of how dynamics correspond to detailed changes in individuals, their number and their interactions.)
  4. Implement the simulation. If any significant element is excessively computationally expensive, consider substituting the element with a behaviourally equivalent analogue (e.g. fractals for protein folding). If the design is still infeasible, change the level of abstraction and redesign the model. A good model should be transparent, intuitive and unambiguous — it should be obvious how it works, what it represents and thus easily extendible. It should be straightforward to monitor and should produce as much data as possible and feasible.
  5. Test the simulation. First, is its representation appropriate? In SC, does each system correspond to something real in the biological system (or does it represent a programmer’s short cut)? Second, does the simulation behaviour correspond to the behaviour you hypothesized, and does it correspond to the behaviour of the biological system under study? If not, check the design and hypothesis. The aim is not to modify the simulation iteratively until it behaves as you wish — the aim is to find a valid hypothesis that can be tested by comparing the behaviour of simulation with reality. For example, a simple electronic calculator may accurately reproduce the ability of our brains to perform addition, but while this aspect of its behaviour may be similar, it provides no explanation of how our brains work. A hypothesis that a network of neurons in our brains can learn to perform addition, tested by a simulation of a neural network that successfully adds numbers together, provides a much more convincing explanation.
  6. Finally, use the simulation to test the hypothetical mechanism. Does its behaviour match the behaviour of the biological system; if you alter the model in specific ways, can the new behaviour (a prediction made by the model) be verified by running new experiments in the biological system; does the model match new previously unseen biological data for the system under study; can the analysis of the inner mechanisms of the model be used as predictors in the biological system; and do new findings about the inner mechanisms in the biological system correspond to the behaviour of the internal systems in the model?
Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License