Albiero Alessandro, Tosatto Silvio C E
Dept. of Biology and CRIBI Biotechnology Centre, University of Padova, V.le G. Colombo 3, 35121 Padova, Italy.
Curr Drug Discov Technol. 2006 Mar;3(1):75-81. doi: 10.2174/157016306776637591.
Modelling of drug targets requires the reliable selection of an accurate and representative structure from large ensembles of alternate models. Statistical potentials developed to discriminate native protein structures generally represent pairwise interactions between atoms, which are less sensitive to local conformational details. The discrimination of local distortions is therefore particularly difficult. Local interaction preferences, expressed through torsion angles, are rarely used, as some controversy exists in the literature regarding their discrimination power. The present study aims to benchmark the efficiency of different implementations of torsion angle propensities for selecting the native structure from ensembles of well-constructed decoys. Several statistical potentials derived from fine-grained discretisations of torsion angle space are constructed and evaluated. Results from a comparison with nine widely used statistical scoring functions show the torsion angle potentials to be more effective in recognising native structures and to improve with the number of torsion angles considered. These data suggest local structural propensities to be important for estimating the overall quality of native-like models.
药物靶点建模需要从大量替代模型集合中可靠地选择准确且具有代表性的结构。为区分天然蛋白质结构而开发的统计势通常表示原子之间的成对相互作用，对局部构象细节不太敏感。因此，区分局部畸变特别困难。通过扭转角表达的局部相互作用偏好很少被使用，因为文献中关于它们的区分能力存在一些争议。本研究旨在对扭转角倾向的不同实现方式从精心构建的诱饵集合中选择天然结构的效率进行基准测试。构建并评估了从扭转角空间的细粒度离散化中导出的几种统计势。与九种广泛使用的统计评分函数比较的结果表明，扭转角势在识别天然结构方面更有效，并且随着所考虑的扭转角数量的增加而改进。这些数据表明局部结构倾向对于估计类天然模型的整体质量很重要。
