BIOLOGICAL MODELLING
Academic work I have done on biological modelling.
METABOLISM-BASED CHEMOTAXIS IN EVOLVED ARTIFICIAL CHEMISTRIES
January 11, 2018
Climbing chemical concentration gradients (chemotaxis) in bacteria has been known and studied for over 100 years, with the basic run-and-tumble mechanism being understood since the late 1800s. Since then, the chemotactic mechanism in bacteria has mostly been studied as a sensory chemical pathway which is independent of other chemical systems in the bacterium, notably the metabolism. Recently Egbert et al. proposed a minimal metabolism-based model of chemotaxis which was able to reproduce an impressive array of experimentally observed chemotactic behaviours in bacteria, demonstrating the simple power of control systems based on internal chemical wellbeing. In reality, however, we know that real metabolic systems are vastly more complex, and consist of autocatalytic networks involving highly specific and complex enzymes. Virgo et al. demonstrated that complex autocatalytic networks could emerge in simple monomer-based chemistries, suggesting that primitive metabolisms might arise more easily than we expect. We extend Egbert's model by considering metabolisms consisting of just such chemistries, showing that a more complex and biologically realistic model of metabolism can be evolved to support chemotaxis. We explore the space of possible evolved control systems, studying properties such as gradient sensitivity and precariousness in the resulting systems. This extends existing work on metabolism-based behaviour, and shows that adaptive behaviour (chemotaxis) can be evolved in an artificial chemistry which is both simple and thermodynamically realistic, suggesting that similar behaviour could evolve in simple proto-cells at the origin of life.
COLLECTIVE SORTING OF CONTINUOUS-SIZE OBJECTS USING A THRESHOLD RULE
February 9, 2018
Ants and other insects are able to sort objects of different types into distinct piles, allowing colony operations to proceed more efficiently. This is done is a decentralized manner, no single insect being in charge. A key principle is stigmergy, which allows for indirect co-operation between agents, and whereby the environment encodes information causing its own re-structuring. In a well-known paper on the subject, Deneubourg proposed a model which is able to sort n different types of object on a grid into distinct clusters of each type using a simple probabilistic rule for picking up or dropping objects in grid cells. We extend this model to account for sorting of objects which vary on a continuous scale, using object size as an example to illustrate the point, although the model is general enough to apply to any continuous quantity which can be measured by agents. We also implement various other real-world aspects, such as a continuous-space environment, more realistic random walks, and collisions between ants. The model is based on a threshold rule for comparing objects, and we find that it is able to sort objects of different sizes, producing piles of roughly equally sized objects even when object sizes are uniformly distributed. The sorting scheme is an emergent property of the threshold rules applied at the local level. This shows that the Deneubourg-style pick up and drop rules are still useful in a more realistic context, and this algorithm could potentially be used by a team of robots to sort objects of varying weights, sizes, or any other measurable quantity.