CDI-Type I: Bridging the Gap Between Next-Generation High Performance Hybrid Computers and Physics Based Computational Models for Quantitative Description of Molecular Recognition

技术摘要：该奖项是根据提交给网络发现与创新倡议的提案，并与刺激竞争性研究的实验项目办公室合作颁发的。该项目以“从数据到知识”和“虚拟组织”为主题。该奖项支持计算机研究和教育，以开发新的极化力场，目的是应用于新药的发现。开发的工具也可能影响新生物材料的发现。分子模拟在新型药物的开发中起着不可或缺的作用。作为实验的补充，计算努力通过筛选具有高结合亲和力、特异性和药理学特性的小分子来加快发现过程。我们感兴趣的基本量是结合亲和力，它与与结合反应相关的自由能变化密切相关。目前的方法采用经验势能函数或力场，表示组成原子/分子物种之间的物理相互作用。相互作用的静电分量以平均场方式处理，在整个结合过程中，分子结构的给定位点上的部分电荷被认为是固定的；量子力学在物理上决定了不同配体和蛋白质构象的电荷分布的不同性质，而量子力学的包含已被证明对准确预测至关重要。PI将通过提出开发下一代经典极化力场的变革性方法来解决药物设计中的这一弱点，以克服当前的限制。挑战在于需要计算能力来解决参数化极化力场的非线性问题，同时结合生物学相关尺度蛋白质-配体结合亲和模拟的结果。随着商用台式机、通用图形处理单元（GPU）和专用现场可编程门阵列（FPGA）中多核架构的出现和集成，混合计算的概念已经进入了高性能计算领域。云计算系统利用这些混合资源来生成更大的数据集，而用户不需要掌握云中的技术基础设施的知识、专业知识或控制权。新兴的云框架还远远不能透明地容纳混合资源。pi将转变开源云计算框架，如Eucalyptus，透明地使用混合资源进行蛋白质配体对接模拟，由基于博弈论策略的智能调度策略驱动。知识推导和使用必须内置在应用程序任务中，并在应用程序任务之间使用，彼此影响？下一步。还需要反馈来自适应地从应用程序任务中驱动新的工作世代。在本项目设想的计算环境中，知识和反馈都需要将人类的知识和洞察力转化为驱动参数搜索的云计算服务。智力优势：从研究的角度来看，本提案中概述的工作将为药物分子极化力场模型的参数化（目标1）和使用极化力场的大型蛋白质配体数据库的蛋白质-配体结合亲和力（目标2）提供自动方法和工具。一个跨校园的云计算系统，透明和智能地使用混合资源，即多核和gpu，在一个统一的，动态适应的工作空间，将支持模拟。该奖项支持教育和推广活动，以促进学生？发现和理解跨学科研究。更广泛的科学和社会影响：就一般科学，特别是建模社区的影响而言，有很大的潜力。对更有效和准确的方法的研究将大大促进药物的发现，并有可能发现新的生物材料。该奖项是根据提交给网络发现和创新倡议的提案，并与实验计划办公室合作，以刺激竞争性研究。该项目以“从数据到知识”和“虚拟组织”为主题。该奖项支持计算机研究和教育，以开发新的计算机模拟工具，以实现分子建模及其相互作用，并潜在地应用于发现新药和发现新的生物材料。这项研究需要大量的计算，以建立分子之间存在的力及其热力学性质的模型。pi将开发一种方法，利用互联网上可用的不同计算资源，使用一种称为？云计算。该奖项支持教育和推广活动，以促进学生？发现和理解跨学科研究。更广泛的科学和社会影响：就一般科学，特别是建模社区的影响而言，有很大的潜力。对更有效和准确的方法的研究将大大促进药物的发现，并有可能发现新的生物材料。

TECHNICAL ABSTRACTThis award is made on a proposal submitted to the Cyberenabled Discovery and Innovation initiative and in partnership with the Office of Experimental Program to Stimulate Competitive Research. This project projects on the From Data to Knowledge, and Virtual Organizations CDI themes. This award supports computational research and education to develop new polarizable force fields with an aim for application to the discovery of new drugs. The tools developed may also impact the discovery of new biomaterials. Molecular simulations play an integral role in the development of novel pharmaceuticals. Complementing experiment, computations strive to expedite the discovery process by screening for small-molecules with high binding affinity, specificity, and pharmacological properties. The fundamental quantity of interest is the binding affinity, and this is rigorously related to the free energy change associated with a binding reaction. Current methods employ empirical potential energy functions, or force fields, representing the physical interactions between the constituent atomic/molecular species. The electrostatic component of the interaction is treated in a mean field manner, with partial charges on given sites of the molecular construct taken to be fixed throughout the binding process; there is no explicit accounting for the differing nature of charge distribution for different ligand and protein conformations as physically dictated by quantum mechanics, whose inclusion has been demonstrated to be vital for accurate predictions. The PI will address this weakness in drug design by proposing transformative approaches to developing next-generation classical polarizable force fields to overcome current limitations. The challenge is the need for computing power to solve the non-linear problem of parameterizing a polarizable force field while simultaneously incorporating results of biologically relevant scale protein-ligand binding affinity simulations. With the emergence and integration of multicore architectures into commodity desktops, general-purpose graphics processing units (GPU), and special purpose field programmable gate arrays (FPGA), the concept of hybrid computing has entered the HPC arena. Cloud computing systems take advantage of these hybrid resources to generate larger data sets, without the users needing to have knowledge of, expertise in, or control over the technology infrastructure in the ?cloud.? Emerging cloud frameworks are far from transparently accommodating hybrid resources. The PIs will transform an open-source cloud computing framework such as Eucalyptus to transparently use hybrid resources for protein-ligand docking simulations, driven by intelligent scheduling policies based on game theory strategies. Knowledge deduction and use have to be built in and used among application tasks, each influencing the other?s next step. Feedback is also needed to adaptively drive new job generations from the application tasks. In the computational environment envisioned in this project, both knowledge and feedback require transforming human knowledge and discernment into cloud computing services that drive the parameter search. Intellectual merit: From the research point of view the work outlined in this proposal will provide automatic methods and tools for the parameterization of polarizable force field models for pharmaceutical molecules (Aim 1) and the protein-ligand binding affinity of large protein-ligand databases using polarizable force fields (Aim 2). A cross-campus cloud computing system that transparently and intelligently uses hybrid resources, i.e., multi-core and GPUs, in a unified, dynamically adaptable workspace, will support the simulations. This award supports educational and outreach activities to advance students? discovery and understanding of interdisciplinary research. Broader scientific and social impacts: There is great potential in terms of impact on the general scientific, and specifically the modeling communities. Research into more efficient and accurate approaches will significantly boost drug discovery and potentially the discovery of new biomaterials.NON-TECHNICAL SUMMARYThis award is made on a proposal submitted to the Cyberenabled Discovery and Innovation initiative and in partnership with the Office of Experimental Program to Stimulate Competitive Research. This project projects on the From Data to Knowledge, and Virtual Organizations CDI themes. This award supports computational research and education to develop new computer simulation tools to enable the modeling of molecules and their interactions with potential application to the discovery of new drugs and the discovery of new biomaterials. The research involves intense computation to create models for the forces that exist between molecules and for their thermodynamic properties. The PIs will develop a method to harness and utilize different computing resources that are available through the internet using a technique called ?cloud computing.? This award supports educational and outreach activities to advance students? discovery and understanding of interdisciplinary research. Broader scientific and social impacts: There is great potential in terms of impact on the general scientific, and specifically the modeling communities. Research into more efficient and accurate approaches will significantly boost drug discovery and potentially the discovery of new biomaterials.