Rooman M, Gilis D
Ingénierie Biomoléculaire, Service de Chimie Organique, Université Libre de Bruxelles, Brussels, Belgium.
Eur J Biochem. 1998 May 15;254(1):135-43. doi: 10.1046/j.1432-1327.1998.2540135.x.
The possibility of defining effective potentials from known protein structures, which are sufficiently accurate to be used for protein-structure-prediction purposes, is investigated. Three types of distance potentials and three types of backbone torsion potentials are defined, based on propensities of amino acid pairs to be separated by a given spatial distance or to be associated to a backbone torsion angle domain. Their differences reside in the way the physical correlations between the amino acids and the conformational states are extracted from the bulk interactions due to the presence of many residues in a protein. For the distance potentials, a physical meaning can be associated to the different definitions, given that some of the potentials favor hydrophobic interactions and others favor interactions between oppositely charged residues. The performance of the different torsion and distance potentials in structure prediction procedures, in particular native-fold recognition and evaluation of protein stability changes upon point mutations, is analyzed. It appears to differ according to the specific proteins and protein environments. In particular, one of the distance potentials performs better than the others for membrane proteins and in protein regions involving charged residues, but less well in other protein regions. Furthermore, the dependence of the potentials on the characteristics of the proteins from which they are derived is analyzed. It is shown that the dependence of the potentials on the length, amino acid composition and secondary-structure content of the proteins from the dataset is either very limited or rather strong, according to the type of potential. The results obtained suggest that the main problem limiting the performance of database-derived potentials is their lack of universality: each potential describes with satisfactory accuracy only the interactions present in certain protein environments.
研究了从已知蛋白质结构定义有效势的可能性，这些结构足够精确，可用于蛋白质结构预测目的。基于氨基酸对被给定空间距离分隔或与主链扭转角域相关联的倾向，定义了三种类型的距离势和三种类型的主链扭转势。它们的差异在于从蛋白质中许多残基的存在所导致的大量相互作用中提取氨基酸与构象状态之间物理相关性的方式。对于距离势，鉴于某些势有利于疏水相互作用，而其他势有利于带相反电荷残基之间的相互作用，不同的定义可以赋予其物理意义。分析了不同扭转势和距离势在结构预测程序中的性能，特别是天然折叠识别和点突变后蛋白质稳定性变化的评估。其表现似乎因特定蛋白质和蛋白质环境而异。特别是，其中一种距离势在膜蛋白和涉及带电荷残基的蛋白质区域中比其他势表现更好，但在其他蛋白质区域中表现较差。此外，分析了势对其衍生蛋白质特征的依赖性。结果表明，根据势的类型，势对数据集中蛋白质的长度、氨基酸组成和二级结构含量的依赖性要么非常有限，要么相当强。所得结果表明，限制基于数据库的势性能的主要问题是它们缺乏通用性：每个势仅能令人满意地描述某些蛋白质环境中存在的相互作用。
