Predicting protein structures with physical petascale molecular simulations

通过物理千万级分子模拟预测蛋白质结构

• 批准号: 1514873
• 负责人: Ken Dill
• 金额: $ 3万
• 依托单位: SUNY at Stony Brook
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2015
• 资助国家: 美国
• 起止时间: 2015-08-01 至 2018-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1514873&HistoricalAwards=false
• 关键词: Predicting; protein; structures; physical; petascale

We have a new computational method (MELD) for predicting protein structures. Unlike others, it is based on physics of atomic interactions. The power of physical methods is that they help us understand how proteins perform their biological actions as catalysts, motors, pumps, transporters and transducers of energy and light. With such algorithms, computers can work hand-in-hand with experiments to provide narratives of the microscopic mechanisms of biology. In the past, physics-based structure prediction was much too computationally costly. The power of MELD is in harnessing weak external knowledge in a way that preserves the power of physical methods (satisfies Boltzmann's Law), and scales up to tackle larger problems of biology than before. Our work with MELD so far has shown that we can use atomistic molecular dynamics to fold 15 out of 20 small proteins to their correct native structures, and to fold 3 small proteins with very high accuracy in the blind test called CASP11 (last summer). No MD simulation can yet fold a protein larger than around 120 amino acids. But, most biologically important proteins are bigger. We have shown that MELD should scale well to larger systems. With this Blue Water allocation we hope to achieve the computational folding of larger proteins. The code we use and the way to implement it is already available in the public GitHub repositories as a plugin to the popular OpenMM molecular dynamics package. With Blue Waters' GPU resources, we hope to bring the power of atomistic MD to folding and mechanisms that can tell some of biology's most interesting stories.

我们有一种新的计算方法（MELD）来预测蛋白质结构。与其他理论不同的是，它是基于原子相互作用的物理学。物理方法的强大之处在于，它们帮助我们理解蛋白质是如何作为催化剂、马达、泵、能量和光的转运体和换能器进行生物活动的。有了这样的算法，计算机可以与实验携手合作，为生物学的微观机制提供叙述。在过去，基于物理的结构预测计算成本太高。MELD的力量在于以一种保留物理方法的力量（满足玻尔兹曼定律）的方式利用薄弱的外部知识，并扩大规模以解决比以前更大的生物学问题。到目前为止，我们与MELD的工作表明，我们可以使用原子分子动力学将20个小蛋白质中的15个折叠成正确的天然结构，并在盲测中以非常高的准确性折叠了3个小蛋白质，称为CASP11（去年夏天）。目前还没有MD模拟可以折叠超过120个氨基酸的蛋白质。但是，大多数生物学上重要的蛋白质都更大。我们已经证明MELD可以很好地扩展到更大的系统。通过这种蓝水分配，我们希望实现更大蛋白质的计算折叠。我们使用的代码和实现方法已经在GitHub公共存储库中作为流行的OpenMM分子动力学包的插件提供。借助Blue Waters的GPU资源，我们希望将原子MD的力量引入折叠和机制，可以讲述一些生物学中最有趣的故事。