High Performance Simulation Libraries for Systems Biology

摘要 细胞过程的定量分析和模拟正在成为一个重要和不可或缺的工具 预测性生物医学研究。不仅如此，模型也正在变得更大、更复杂和 包括组织和整个器官的多细胞和多尺度建模等领域。性能 可以通过使用计算集群或云来改进，但交互计算受到严格限制 使用这种基础设施。此外，并不是每个人都可以使用超级计算或云访问。在这 我们希望专注于开发高性能的模拟器，特别是libRoadRunner和特定的 基于模拟计算的硬件。LibRoadRunner的独特之处在于它是唯一一个 使用LLVM编译标准SBML。LLVM是一种正在起步的后端机器代码生成技术 供注重性能的软件开发人员广泛使用。使用这项技术，我们已经实现了 重要的是，在某些情况下，与现有的模拟器相比，模拟时间的速度提高了几个数量级。 我们的支持信强调了这一新技术实现的一些重大改进 技术我们已经证明，libRoadRunner的性能与Native相当(在95%以内 已编译的解决方案。换句话说，我们已经达到了在普通台式计算机上进行计算的极限。 在这个提案中，我们将继续改进libRoadRunner，并实现一种全新的方式来实现 蜂窝模拟，这将导致模拟性能的额外数量级改进。 我们的建议是与达特茅斯的拉胡尔·萨佩什卡合作，他开创了基于硬件的模拟 被称为细胞形态计算的技术。作为这项提议的一部分，达特茅斯小组将开发一种新的 一代基于硬件的超高速生化模拟器。我们将开发SBML和MATLAB 翻译器，可以将模型直接翻译成可编程的高性能细胞形态硅芯片。 除了这种高度创新的方法外，我们还将利用这种硬件来实现新的方法 到细胞过程的交互式实时建模，包括对时间和时间的完全控制 入侵。拟议工作的一小部分将包括对现有基础设施的一般维护，如 并每年组织一次建模和培训研讨会。

Abstract Quantitative analysis and simulation of cellular processes is becoming an important and indispensible tool for predictive biomedical research. Not only that, but models are becoming larger, more sophisticated and encompassing areas such as multicellular and multiscale modeling of tissues and whole organs. Performance can be improved by using compute clusters or the cloud, however interactive computing is severely restricted using this kind of infrastructure. Moreover supercomputing or cloud access is not available to everyone. In this proposal we wish to focus on developing high performance simulators, notably libRoadRunner and specific hardware based on analog computing. libRoadRunner is unique in that it is the only cellular simulator that compiles standard SBML using LLVM. LLVM is a backend machine code generation technology that is starting to be widely used by performance conscious software developers. Using this technology we have achieved significant, in some cases orders of magnitude speed ups in simulation times compared to existing simulators. Our letters of support highlight some of the major improvements that have been realized by this new technology. We have shown that the performance of libRoadRunner is on par (within 95%) with natively compiled solutions. In other words we have reached the limit to computing on ordinary desktop computers. In this proposal we will continue to improve libRoadRunner and implement a completely novel way to carry out cellular simulations which will lead to additional orders of magnitude improvements in simulation performance. Our proposal is to work with Rahul Sarpeshkar at Dartmouth who has pioneered hardware based simulation technologies called cytomorphic computing. As part of this proposal, the Dartmouth group will develop a new generation of ultra-high-speed hardware-based biochemical simulators. We will develop SBML and MATLAB translators that can translate a model directly into programmable high-performance cytomorphic silicon chips. In addition to this highly innovative approach we will also exploit this hardware to implement novel approaches to interactive real-time modeling of cellular processes that includes the complete control of time and temporal incursion. A small part of the proposed effort will include general maintenance of our existing infrastructure as well as organizing an annual workshop on modeling and training.