ABSTRACT MODELLING OF EVOLUTION OF NEURAL CIRCUITS WITH POPULATION-BASED TRAITS ENCODING

In distantly related animals, nervous systems evolved to produce functionally similar behavior. The basic physiology of a single neuron is also similar. However, the anatomical structures and neural circuits display striking diversity due to a variety of evolutionary paths that produced them. Uncovering the evolutionary history of neural circuits using biological data is extremely difficult. Computational modelling may help detect some of the guiding principles of this process.
I present a prototypical pipeline to simulate a large-scale evolution of neural circuits. It utilizes neuronal subclass-based properties encoding and natural selection of circuits based on output activity. Each circuit receives an input signal in the form of spiking activity of afferent neurons. The selection process evaluates the activity of the efferent neurons generated in response to the input. The interneurons participate in the transformation of the input signal into an output, but their activity does not affect selection directly. The circuits that generate the most “adaptive” output leave the offspring to the next generation, where the selection process repeats. The offspring inherit not properties of individual neurons, but subclasses of neurons with properties defined statistically. Mutations that may occur between the generations affect not individual neurons but entire subclasses. This way, generalized statistical descriptions of circuits are being selected based on the activity patterns of the efferent neurons.
Initial tests demonstrated that circuits indeed evolve in this paradigm. Properties and connection patterns of subclasses change under selection pressure to improve the output activity, increasing circuits’ fitness over generations.