The Digital Fly Brain

目前，几个雄心勃勃的、大规模的、数十亿英镑的研究项目正在进行中，这些项目旨在了解人类大脑。在欧洲，人类大脑项目致力于通过开发多尺度、多层次的人类大脑模型，整合来自众多不同研究项目的数据，加速大脑研究--1000亿神经元建模和模拟挑战。在美国，大脑计划旨在重建整个神经回路的神经活动的完整记录-1000亿个神经元记录挑战。这些显然是巨大的，但有价值的挑战，可以从理解更小但足够复杂的大脑的神经计算原理中受益。果蝇大脑已经成为研究神经计算和将大脑结构与功能联系起来的最受欢迎的模型生物之一。许多在哺乳动物大脑中表达的基因和蛋白质在果蝇的基因组中也是保守的。值得注意的是，果蝇能够进行许多复杂的非反应性行为，这些行为由仅包含约100，000个神经元的大脑控制。苍蝇的大脑和它的行为之间的关系可以通过实验来探索，使用强大的遗传技术工具包来操纵苍蝇的神经回路。精确记录果蝇神经元对刺激的反应和绘制果蝇神经系统神经元和突触的新实验方法提供了大量关于果蝇神经连接图及其对感觉刺激的处理的有价值的数据。这些特征加上越来越多的伦理和经济压力，以减少在研究中使用哺乳动物，解释了对基于果蝇的大脑模型越来越感兴趣，不仅要了解感知，感知和神经计算，而且还要获得可能为我们提供信息的机制见解努力解决人类神经退行性疾病，如阿尔茨海默病。该项目旨在设计，实施和实验评估一个潜在的变革性开源苍蝇大脑模拟平台，该平台能够模拟构成成年果蝇大脑的约135，000个神经元。这种计算基础设施将基于最近建立的图形处理器单元（GPU）支持的神经内核软件平台。该模块化仿真平台将整合有关果蝇大脑的所有知识，作为一组相互连接的仿真模块，描述了大约41个本地处理单元（LPU），6个集线器及其相互连接的操作，部分通过详细的EM成像研究阐明。该模拟平台将用于开发和验证第一个模型草案，该模型将结合最先进的神经元生物物理和/或功能模型以及最新发布的突触连接图。主要重点将是开发早期视觉系统（视网膜，板，髓质）和早期嗅觉系统（OSN，触角叶，蘑菇体，侧角）的详细模型。这些模型将整合视觉和嗅觉系统的完整模型。大脑模拟平台将首次实现对苍蝇大脑模型神经回路及其连接模式的隔离和集成仿真（例如，感觉和运动系统）以及苍蝇神经系统的其他部分。使用Neurokernel仿真平台，可以足够快地生成数据，使研究人员能够在实验运行时在线比较和调整虚拟神经元的输入输出特性。

Several highly ambitious, large-scale, billion-pound research projects that aim to understand the human brain are currently under way. In Europe, The Human Brain Project is focused on accelerating brain research by integrating data available from a multitude of disparate research projects through the development of a multi-scale, multi-level model of the human brain - the 100 billion neurons modelling and simulation challenge. In the US, The Brain Initiative aims to reconstruct the full record of neural activity across complete neural circuits - the 100 billion neurons recording challenge. These are clearly huge, but worthy challenges that can benefit from an understanding of the principles of neural computation of much smaller yet sufficiently complex brains. The fruit fly brain has become one of the most popular model organisms to study neural computation and for relating brain structure to function. Many of the genes and proteins expressed in the mammalian brain are also conserved in the genome of the fruit fly. Remarkably, the fruit fly is capable of a host of complex nonreactive behaviors that are governed by a brain containing only ~100,000 neurons. The relationship between the fly's brain and its behaviors can be experimentally probed using a powerful toolkit of genetic techniques for manipulation of the fly's neural circuitry. Novel experimental methods for precise recordings of the fly's neuronal responses to stimuli and for mapping neurons and synapses in Drosophila nervous system have provided access to an immense amount of valuable data regarding the fly's neural connectivity map and its processing of sensory stimuli. These features coupled with the growing ethical and economic pressures to reduce the use of mammals in research, explain the growing interest in Drosophila-based brain models, not only to understand sensing, perception and neural computation, but also to gain mechanistic insights that may inform our efforts to address neurodegenerative diseases, such as Alzheimer's disease, in humans. This project aims to design, implement and experimentally evaluate a potentially transformative open-source fly brain simulation platform capable of simulating ~135,000 neurons that make up the adult Drosophila brain. This computational infrastructure will be based on the recently established Graphic Processor Units (GPU)-enabled Neurokernel software platform. The modular simulation platform will integrate all knowledge about the Drosophila brain as a set of interconnected simulation modules which describe the operation of about 41 Local Processing Units (LPUs), six hubs and their interconnections, partly elucidated by detailed EM imaging studies. The simulation platform will be used to develop and validate a first draft model that incorporates the most advanced biophysical and/or functional models of the neurons and the latest published synaptic connections maps. The main focus will be on developing detailed models of the early visual system (retina, lamina, medulla) and of the early olfactory system (OSNs, antennal lobe, mushroom body, lateral horn). These models will integrate complete models of the visual and olfactory systems. The brain simulation platform will enable for the first time the isolated and integrated emulation of fly brain model neural circuits and their connectivity patterns (e.g., sensory and locomotion systems) and other parts of the fly's nervous system on clusters of GPUs. Using the Neurokernel simulation platform it will be possible to generate data sufficiently fast to enable researchers to compare and tune the input-output characteristics of virtual neurons on-line, while the experiment is running.