ENERGY SMOOTHING METHODS FOR CONFORMATIONAL SEARCH

生物聚合物计算机建模的双重挑战是在能量学描述中实现化学准确性，以及为大型柔性系统充分寻找构象空间。本课题旨在改进和扩展势能平滑技术在构象搜索中的应用。目前的搜索方法，如模拟退火，是结构生物学中许多正在进行的研究的关键。通过对比原始表面光滑的经过数学变换的势能面进行采样，可以在很大程度上克服搜索问题。在这个项目中，来自Scheraga关于扩散方程方法的工作和Straub关于高斯密度退火的工作的思想构成了新的势平滑搜索（PSS）算法的基础，在构象搜索应用中表现出更大的收敛范围。导出了粗粒度平滑能量函数和溶剂化模型的解析表达式。结合更好的局部搜索启发式和各向异性原子概率分布的组合方法将优于广泛使用的退火协议。对于许多系统，PSS方法可以在单个确定性计算中产生全局能量最小值。一种相关的算法，构象扫描，发现了一组结构，代表了跨越有机分子、肽和小蛋白质潜在表面的独特低能盆地。这些平滑算法将应用于构象搜索和结构优化、小分子晶体的包装、柔性配体与蛋白质结合位点的对接以及分子构象的表征和聚类等问题。重点将放在建立螺旋跨膜蛋白的模型，如糖蛋白、细菌视紫红质、视紫红质和g蛋白偶联受体家族。跨膜螺旋之间的相互作用是许多整体膜蛋白结构和功能的主要决定因素。充分了解这些体系需要原子分辨率结构知识，这很难通过核磁共振方法的晶体学获得。PSS方法为这些重要的生物系统建模提供了一种适当而有效的搜索范式。

The twin challenges in computer modeling of biopolymers are achieving chemical accuracy in the description of energetics, and adequately searching conformational space for large flexible systems. This project proposes to improve and extend the application of potential energy smoothing techniques for conformational search. Current search methods, such as simulated annealing, are critical to much of the ongoing research in structural biology. The search problem can be largely overcome by sampling on mathematically transformed potential energy surfaces which are smoother than the original surfaces. In this project, ideas derived from work by Scheraga on the Diffusion Equation Method and Straub s work on Gaussian Density Annealing form the basis of new Potential Smoothing Search (PSS) algorithms exhibiting a much larger range of convergence in conformational search applications. Analytical expressions for course-grained, smoothable energy functions and solvation models are derived. Incorporation of better local search heuristics and anisotropic atomic probability distributions should result in a combined method superior to widely used annealing protocols. For many systems, PSS methods can produce the global energy minimum in a single deterministic calculation. A related algorithm, Conformational Scanning, finds sets of structures representing unique low energy basins spanning the potential surfaces of organic molecules, peptides and small proteins. These smoothing algorithms will be applied to problems in conformational search and structure refinement, packing of small molecule crystals, flexible ligand docking to protein binding sites, and the characterization and clustering of molecular conformations. Emphasis will be placed on developing models for helical transmembrane proteins such as glycophorin, bacteriorhodopsin, rhodopsin and the family of G-protein coupled receptors. Interactions between transmembrane helices are central determinants of structure and function for many integral membrane proteins. Full understanding of these systems requires atomic resolution structural knowledge which is difficult to obtain via crystallographic of NMR methods. PSS methods represent an appropriate and efficient search paradigm for modeling these biologically important systems.