Reconstruction of brain networks by algorithmic amplification of morphometry data

Abstract

The peculiar shapes of neurons have been fascinating since their discovery in the late 1800’s, when the Golgi impregnation technique established circuits of neurons and glia to be the hallmark of brain organization. Although only a small fraction of all brain cells have been carefully measured, it is a natural complement to reductionism to try to logically re-construct brain circuitry using its basic units. This paper describes a computational strategy to produce virtual models of biological neural networks detailed at the micron level. Our algorithm uses cylindrical primitives in a virtual reality environment, and assembles them according to basic growth rules. Morphometric parameters (e.g., the axonal stem’s diameter) are measured from libraries of experimentally traced neurons and stored as statistical distributions. When the algorithm uses a parameter to generate a neuron (e.g. when an axon stems from the soma and its initial diameter needs to be determined), a value is stochastically sampled from the statistical distribution. This procedure can produce a large number of non-identical virtual neurons, whose morphological characteristics are statistically equivalent to those of the original experimental neuron. Thus, an amplification of morphometry data is achieved. This stochastic and statistical approach is highly efficient, allowing the creation of large-scale, anatomically accurate neural networks.