Modelling Speciation

Modelling Speciation



This "speciation" model uses populations of agents which possess various traits (600+ agents in small population tests, 1600 in large populations). Agents are abstractions of family groups, packs, etc. in some virtual ecosystem.

Get the current model here speciation model.

Agents breed with near neighbours, passing on their traits. Breeding ranges and other factors can be varied. By introducing weakly enforced, population-wide preferences for mate-selection, speciation occurs without any other selection pressures. Speciation in this case implies the evolution of collections of agents which share similar traits (and are therefore members of the same species) but that these traits are divergent from those of other agents (who are therefore in other species).

Typical trials using smaller populations partially converge after a few hundred cycles (from initial genetic variation of 150 types to 10-15 types) and completely converge after 2000+ cycles (to stable states of 5-10 gene types).

The speciation model uses simple graphics to show traits/characteristics of individuals. The image (left) shows an individual with 4 trait values...

  • foreground shape: target
  • foreground color: yellow
  • background shape: orbit
  • background color: red

The model shows a 2D world of agents and animates the changes to their traits as they occur. When the model starts up agents are given random trait values. As the model runs (and depending on the type of phasing & selection used) agents (chosen at random) adopt one trait from one of their potential breeding candidates (usually one of their neighbours) simulating a breeding cycle.

Over time agents are allowed to exert a preference for mates with some similarity in traits to themselves (ie: a weak sexual selection pressure for some similarity). Over time this causes the model to converge into localised clusters of similar individuals who no longer breed (swap traits) with neighbouring clusters. We believe this shows a type of speciation - our analysis is on-going.

4 screen-shots of the model world over time (1st one is after setup, last one almost converged)...




4 screen-shots of a model with "walls" (last one fully converged)...



NB: the individuals in the world are abstract, more realistically representing extended family groups or packs. In some analysis it may be useful to consider genetic drift occurring within these clusters.



Full screen-shot of model. Running standard "local" selection with no "walls"...


This page and the work are still under development. For mode details send us an email.