AbstractThe intricate morphology of neurons has fascinated since the dawn of neuroscience, and yet, it is hard to synthesize them. Current algorithms typically define a growth process with parameters that allow matching aspects of the morphologies. However, such algorithmic growth processes are far simpler than the biological ones. What is needed is an algorithm that, given a database of morphologies, produces more of those. Here, we introduce a generator for neuron morphologies that is based on a statistical sampling process. Our Reversible Jump Markov chain Monte Carlo (RJMCMC) method starts with a trivial neuron and iteratively perturbs the morphology bringing the features close to those of the database. By quantifying the statistics of the generated neurons, we find that it outperforms growth-based models for many features. Good generative models for neuron morphologies promise to be important both for neural simulations and for morphology reconstructions from imaging data.
Flow-based generative models for Markov chain Monte Carlo in lattice field theory