Bayesian methods for protein structure calculation from sparse, heterogenous and lowquality data

蛋白质在活细胞中通过它们折叠成的特定三维结构来执行各种功能。确定蛋白质结构的方法包括X射线晶体学、核磁共振光谱学和电子显微镜。随着数据质量和数量的下降，结构确定往往成为一个通过或失败的问题。即便如此，数据仍然可能提供信息。该项目的目的是开发计算工具，用于从传统上被认为不足以达到原子分辨率的实验数据中计算蛋白质结构。这些工具应该自动运行，只需要最少的人为干预。数据中的缺陷将由先前的结构知识补偿。关于结构的其余不确定性需要得到充分体现。这需要对与数据和先验知识同等兼容的替代构象进行采样，并包括对其精确度和可能性的量化。贝叶斯概率理论提供了最佳的数学框架，开发这些工具。它使噪声和不完整的数据的无偏分析，整合来自不同来源的结构信息，并建立了一个推理机制，估计参数的不确定性和缺失信息。

Proteins carry out diverse functions in the living cell by means of a specific three-dimensional structure into which they fold. Methods to determine protein structures include X-ray crystallography, nuclear magnetic resonance spectroscopy, and electron microscopy. With decreasing data quality and quantity, structure determination often becomes a matter of pass or fail. Even so the data may still be informative. The aim of this project is to develop computational tools for calculating protein structures from experimental data that traditionally have been considered insufficient for atomic resolution. These tools should operate automatically and require only minimal human intervention. Deficiencies in the data will be compensated for by prior structural knowledge. The remaining uncertainty about the structure needs to represented adequately. This requires the sampling of alternative conformations which are equally compatible with both data and prior knowledge and includes the quantification of their precision and likelihood. Bayesian probability theory provides the optimal mathematical framework to develop these tools. It enables the unbiased analysis of noisy and incomplete data, integrates structural information from diverse sources and furnishes an inference machinery to estimate parameter uncertainties and missing information.