The functions that a network of neurons may perform is shaped by the map of connections between the neurons. Understanding the principles that govern the architecture of these networks of connections, known as "connectomes," is crucial for advancing our knowledge of brain development and architecture, how learning occurs (when these connections change), and the behavior these networks direct. Researchers at Elad Schneidman’s group are developing “generative models” for the connectomes of networks in the brains of different organisms: the olfactory bulb of zebrafish, the visual cortex of mice, and the complete neural network of the C. elegans worm.
They discovered that in all three cases, AI-driven models that rely on a surprisingly small number of simple biological and physical features, accurately replicate a wide range of properties observed in these circuits. Specifically, the models accurately predicted various aspects, such as the existence and strength of individual synapses (connections between neurons), the frequency of sub-network patterns and more. Moreover, simulating synthetic circuits of the olfactory bulb of zebrafish generated by the model demonstrated similar responses to olfactory cues as the real one.
In summary, this study suggests that surprisingly simple design principles underlie the structure of connectomes across different systems and species. It also provides a new computational framework for analyzing connectomes and understanding the link between structure and function in neural circuits.