Bayesian estimation of network connectivity and motifs

点击翻译按钮获取中文摘要

Abstract The overarching goal of this proposal is to learn how large groups of neurons interact in a network to perform computations that go beyond the individual ability of each cell. Our working hypothesis is that emergent behavior in neural networks results from their organization into a hierarchy of modular sub-networks or motifs, each performing simpler computations than the network as a whole. This theoretical framework suggests that our understanding of neural networks will advance if we can reliably measure network connectivity, detect recurring motifs, elucidate the computations they perform, and reveal how these smaller modules are combined into larger networks capable of performing increasingly complex computations. To advance the field forward we will: (a) develop novel system identification methods for cortical networks based on dynamical, two- photon imaging data. Our methods will use a Bayesian formulation that incorporates prior constraints on network topology, sparsity of synaptic connections, and cell type, derived from published, experimental data; (b) advance graph theoretic methods to identify patterns of connectivity among subsets of neurons which appear at rates higher than chance; (c) will use extensive in-vivo and in-vitro methods to validate our techniques. The work will deliver transformative software tools for Bayesian inference of network connectivity from functional data; it will yield a catalog of elementary cortical motifs of excitatory and inhibitory cells that will shed light on the wiring of the cortical circuitry; and it will generate the first database combining functional calcium imaging data with “ground truth” estimates of direct synaptic connectivity. Altogether, the proposed work will make available much needed analytical tools and databases to support a wide range of studies under the BRAIN initiative.