To B or not to B: a question of resolution?
To B or not to B: a question of resolution?
复制标题
DOI：
10.1107/s0907444911028320
复制
发表时间：
2012-04
期刊：
影响因子：
--
通讯作者：
Merritt EA
中科院分区：
文献类型：
--
作者：
Merritt EA
基于分辨率的简单经验法则不足以确定大分子结构模型中原子位移的最佳处理。选择使用各向同性B因子、各向异性B因子、TLS模型或三者的某种组合应通过模型细化的统计分析进行验证。在选择和改进任何晶体学结构模型时，在提取尽可能详细的信息的愿望和描述不超过观察数据所证明的必要性之间存在紧张关系。更复杂的模型不一定是更好的模型。因此，重要的是要验证参数的选择以及验证其细化值。一个经常性的任务是选择最好的模型来描述每个原子关于其平均位置的位移。在原子分辨率下，可以选择六个模型参数（“热椭球”）来描述每个原子的位移。在中等分辨率下，通常每个原子最多只能用一个模型参数来描述同一事物（“B因子”）。在非常低的分辨率下，不能证明每个原子使用甚至一个参数是合理的。此外，这方面的结构可以更好地描述一个显式模型的整体位移，其中最常见的是平移/天平动/螺旋（TLS）形式主义，而不是通过分配一些数量的参数，每个原子单独。人们可以通过在同一个模型中包括两种处理来回避原子位移参数和TLS描述之间的选择，但这并不总是统计上合理的。对于特定分辨率下的特定结构精修，哪种处理方法是最佳的选择可以通过模型参数与观察次数之比的一般考虑以及通过特定的统计数据（例如汉密尔顿R-因子比检验）来指导。
A simple rule of thumb based on resolution is not adequate to identify the best treatment of atomic displacements in macromolecular structural models. The choice to use isotropic B factors, anisotropic B factors, TLS models or some combination of the three should be validated through statistical analysis of the model refinement. In choosing and refining any crystallographic structural model, there is tension between the desire to extract the most detailed information possible and the necessity to describe no more than what is justified by the observed data. A more complex model is not necessarily a better model. Thus, it is important to validate the choice of parameters as well as validating their refined values. One recurring task is to choose the best model for describing the displacement of each atom about its mean position. At atomic resolution one has the option of devoting six model parameters (a ‘thermal ellipsoid’) to describe the displacement of each atom. At medium resolution one typically devotes at most one model parameter per atom to describe the same thing (a ‘B factor’). At very low resolution one cannot justify the use of even one parameter per atom. Furthermore, this aspect of the structure may be described better by an explicit model of bulk displacements, the most common of which is the translation/libration/screw (TLS) formalism, rather than by assigning some number of parameters to each atom individually. One can sidestep this choice between atomic displacement parameters and TLS descriptions by including both treatments in the same model, but this is not always statistically justifiable. The choice of which treatment is best for a particular structure refinement at a particular resolution can be guided by general considerations of the ratio of model parameters to the number of observations and by specific statistics such as the Hamilton R-factor ratio test.