Computer Methods for Physiological Problems

描述(申请人提供)：这个项目的主要目标是增强神经元的并行能力；这是由研究人员开始提出和研究的关于健康和疾病中大脑功能的大规模生物物理现实模型所推动的。神经元是我们为OS X、MS Windows和Unix/Linux开发并免费提供的神经元和神经电路的开源模拟器。它简化了数值方法和编程方面的非专业人员的神经模型的创建和分析。神经元在工作站、集群和大规模并行的超级计算机上运行，模型属性可以包括但不限于复杂的分支形态、多种通道类型、不均匀的通道分布、离子扩散、细胞外场、电子仪器和人工尖峰细胞。越来越多的神经科学家在世界各地的实验室使用它来研究一系列疾病，如癫痫、多发性硬化症和学习和记忆障碍的正常和异常大脑功能的基础机制，以及药物和脑深部刺激(DBS)等治疗干预措施如何影响这些机制。他们正在解决的问题涉及越来越大的解剖和生物物理复杂性，这使得执行速度受到重视，但硬件串行执行性能在五年多的时间里一直保持不变。神经元已经支持尖峰耦合网络、间隙连接和单个细胞的分布式电缆模型的线程和MPI风格的并行性，但我们已经确定了三个仍可实现重大改进的领域。第一个是GPU并行化，这将使具有复杂膜属性的风格化细胞的网络模型实现超过一个数量级的速度提升。第二是神经元微分-代数方程解算器的并行化，这将使涉及细胞外场、反应-扩散和细胞间触觉相互作用的模型能够利用多核和并行集群架构。三是实现模型设置的多线程并行化。帮助 为了评估和完善这些新功能，我们将实现DBS模型，由于其已证明的和潜在的临床实用价值，这是许多实验和计算研究的重点。该项目的传播和支持部分包括：通过NeuroML 2增强神经元与其他模拟器的互操作性；软件维护，以确保用户能够继续在OS X、MS Windows、Unix/LINUX的所有组合下运行神经元，目前32位和位处理器架构上仍在使用各种版本的Python；通过电子邮件和神经元论坛积极支持用户；创建新的文件和发表文章和报告；举办课程和组织神经元用户会议。

DESCRIPTION (provided by applicant): The primary goal of this project is to enhance the parallel capabilities of NEURON; this is motivated by the large scale of biophysically realistic models of brain function in health and disease that investigators are beginning to propose and study. NEURON is an open source simulator of neurons and neural circuits that we have developed and provide freely for OS X, MS Windows, and UNIX/Linux. It simplifies the creation and analysis of neural models for nonspecialists in numerical methods and programming. NEURON runs on workstations, clusters, and massively parallel supercomputers, and model properties may include, but are not limited to, complex branching morphology, multiple channel types, inhomogeneous channel distribution, ionic diffusion, extracellular fields, electronic instrumentation, and artificial spiking cells. Growing numbers of neuroscientists in laboratories around the world are using it to investigate mechanisms that underlie normal and abnormal brain function in a wide range of disorders, e.g. epilepsy, multiple sclerosis, and disorders of learning and memory, as well as how these are affected by therapeutic interventions such as medications and deep brain stimulation (DBS). The problems they are addressing involve ever greater anatomical and biophysical complexity, which puts a premium on execution speed, yet hardware serial execution performance has remained unchanged for more than half a decade. NEURON already supports thread and MPI styles of parallelism for spike coupled networks, gap junctions, and distributed cable models of individual cells, but we have identified three areas where large improvements are still achievable. The first is GPU parallelization, which will allow network models of stylized cells with complex membrane properties to achieve more than an order of magnitude speed increase. The second is parallelization of NEURON's differential-algebraic equation solver, which will enable models involving extracellular fields, reaction-diffusion, and ephaptic interactions between cells to take advantage of multicore and parallel cluster architectures. The third is to enable multithreaded parallelization of model setup. To help evaluate and refine these new features, we will implement a model of DBS, a focus of much experimental and computational research because of its demonstrated and potential clinical utility. The dissemination and support component of this project includes: enhancing NEURON's interoperability with other simulators via NeuroML 2; software maintenance to ensure that users continue to be able to run NEURON with or without Python under all combinations of OS X, MS Windows, UNIX/Linux, with the various versions of Python still in use, on current 32- and 64-bit processor architectures; active support of users through email and the NEURON Forum; creation of new documentation and publication of articles and reports; presenting courses and organizing meetings of NEURON users.