Relating structure and function in synapse-level wiring diagrams

点击翻译按钮获取中文摘要

Project summary: Modern electron-microscopy (EM) imaging and analysis methods now permit the comprehensive reconstruction of all neurons and synapses in large volumes of brain tissue or the entire brains of individual organisms. However, relating this structure to function is difficult. The rapidly increasing scale of these datasets requires the develop- ment of new quantitative techniques to address this challenge. This proposal describes a combined data analysis and modeling approach that is informed by large-scale EM datasets collected by our experimental collaborators. The methods we will develop extend the state of the art by incorporating multiple sources of information about neuronal connectivity and function to explain structure in EM wiring diagrams. They also leverage recent advances in recurrent neural network optimization to use this structure to constrain models of neural dynamics. Our aim is both to develop general and scalable techniques to be used on the latest generation of datasets, as well as apply these techniques to specific scientific questions about the organization of the Drosophila mushroom body, which is a primary target of current reconstruction efforts. Specific aims of the project include a number of subgoals, starting with the development of techniques to determine the organizing principles of neuronal wiring given a connectivity graph defined by an EM dataset. Unlike standard methods, we aim to leverage multiple modalities of information; for instance, connectivity, cell types, functional data, spatial location, and synaptic weights, to perform this inference. Next, we will perform an analysis of the mushroom body of the adult Drosophila melanogaster brain, a center for associative learning in insects. This analysis will both inform the development of our methods and also address fundamental scientific questions about the nature of stimulus representations in mushroom body Kenyon cells and the circuitry involved in learning. Strong parallels between the organization of the mushroom body and the mammalian cerebellum suggest that these efforts will lead to generalizable insights. Finally, we will integrate structural information with modeling of neural dynamics. We will characterize to what extent structure can be used to build well-constrained models, validating our approaches on datasets that involve characterization of connectivity through EM and recording of neural activity through calcium imaging. The proposed methods will be of interest for researchers working across many model organisms for which EM reconstruction efforts have been completed or are currently underway. We expect that the methods will provide a template for integrating structural information into modeling efforts across these varied systems.