Bayesian analysis of structural shape and conformation

描述(申请人提供)：对结构差异和共性的准确分析对于理解生物大分子的结构、功能和进化是至关重要的。在过去的40年里，结构分析方法一直依赖于生物物理上不现实的限制性最小二乘准则来寻找最优叠加。通过开发可以利用强大的最大似然(ML)和贝叶斯技术的结构变化的概率模型，这项拟议的工作将极大地扩展我们准确叠加、排列和分析结构构象的能力。这项工作的具体目标是(1)发展用于叠加结构构象的贝叶斯模型和理论，(2)发展基于多结构比对的最大似然和贝叶斯模型和理论，以及(3)开发和分发实现这种最大似然和贝叶斯结构分析的计算工具。最大似然法和贝叶斯结构分析将提供许多不同于当前最小二乘和其他特别方法的优势，包括(1)对估计参数的解的不确定性的直接估计，(2)对结构数据中的不确定性的优雅处理，(3)自然结合不同类型的先前结构和分子信息，(4)容易检查结构变化和演化的复杂模型，(5)对复杂结构假设的严格测试，以及(6)对丢失的结构数据的自然处理。虽然我们专注于大分子的构象，但这里提出的方法具有广泛的数学普遍性，不仅将影响分子结构生物学，还将影响范围异常广泛的科学领域，包括任何比较物体形状和构象的领域。这项工作得出的结果将适用于任何可以表示为多维空间中一组笛卡儿点的实体，无论研究中的特定结构是蛋白质、头骨、核磁共振扫描、地质地层，甚至是人类个体的心理特征。

DESCRIPTION (provided by applicant): Accurate analysis of structural differences and commonalities is of fundamental importance for understanding the structure, function, and evolution of biological macromolecules. For the past 40 years, structural analysis methods have relied on the biophysically unrealistic and restrictive least-squares criterion to find optimal superpositions. By developing probabilistic models of structural change that can take advantage of powerful maximum likelihood (ML) and Bayesian techniques, this proposed work will greatly expand our abilities to accurately superposition, align, and analyze structural conformations. The specific aims of this work are (1) to develop Bayesian models and theory for superpositioning structural conformation, (2) develop ML and Bayesian models and theory for multiple structure-based alignment, and (3) develop and distribute computational tools that implement this ML and Bayesian structural analysis. ML and Bayesian structural analysis will provide many distinct advantages over current least-squares and other ad hoc methods, including (1) straight- forward estimates of the uncertainty in the solutions of estimated parameters, (2) elegant handling of uncertainty in structural data, (3) natural incorporation of disparate types of prior structural and molecular information, (4) easy examination of complex models of structural change and evolution, (5) rigorous testing of complex structural hypotheses, and (6) natural handling of missing structural data. While we concentrate specifically on the conformations of macromolecules, the methods proposed herein have broad mathematical generality and will impact not only molecular structural biology but also an unusually wide range of scientific fields, including any that compare the shapes and conformations of objects. The results developed from this work will be applicable to any entity that can be represented as a set of Cartesian points in a multi-dimensional space, whether the particular structures under study are proteins, skulls, MRI scans, geological strata, or even psychological profiles of human individuals.