Fold-recognition detects an error in the protein data bank
Fold-recognition detects an error in the protein data bank
复制标题
DOI：
10.1093/bioinformatics/18.10.1391
复制
发表时间：
2002-10-01
期刊：
影响因子：
5.8
通讯作者：
Fischer, D
中科院分区：
文献类型：
--
作者：
Bujnicki, J;Rychlewski, L;Fischer, D
通过X射线晶体学确定蛋白质的3D结构，需要将氨基酸序列拟合到电子密度图中。如果图谱分辨率低，显示的侧链数据很少，这可能是一项困难而耗时的任务。在这里，我们展示了如何折叠识别，使我们能够检测到一个错误，在最近存放的PDB条目，并提出了一种方法，可以在拟合的序列到一个低分辨率的电子密度图的援助。所描述的过程是在蛋白质结构预测基准测试实验LiveBench（Bujnicki等人，2001年a）。蛋白质结构预测的目的是为未知结构的目标蛋白质序列生成近似的3D结构模型。特别地，折叠识别或穿线针对与任何已知结构的蛋白质具有很少或没有显著序列相似性的那些靶标（Fischer等人，1996年）。LiveBench连续基准测试程序通过提交新发布的PDB条目的序列作为预测目标来评估自动折叠识别服务器的性能，这些PDB条目与先前发布的蛋白质没有明显的序列相似性。在从参与服务器收集预测模型之后，通过将它们与实验结构进行比较来评估预测。在最近考虑的靶序列中，有许多新发布的PDB条目1 kc 9的链，它描述了细菌D的核糖体大亚基的晶体结构。抗辐射剂（Harms，2001）。这是一个只有Cα的结构，分辨率为3.1 A。在这里，我们专注于M链，其对应于核糖体蛋白L18的残基2-114的结构，在本文中称为1 kc 9 M。所有的折叠识别方法（Bujnicki等人，2001 b）参与LiveBench（包括迭代PSI-BLAST检索），非常有信心地表明，
Determining the 3D-structure of a protein by means of X-ray crystallography, requires the fitting of the amino acid sequence into an electron-density map. This can be a difficult, time-consuming task if the map is of low resolution, showing little data for the side-chains. Here we show how fold recognition allowed us to detect an error in a recently deposited PDB entry, and propose a method that can be of aid in the fitting of a sequence into a lowresolution electron-density map. The described procedure is the result of an interesting anomaly detected during the protein structure prediction benchmarking experiment, LiveBench (Bujnicki et al., 2001a). Protein structure prediction is aimed at generating approximate 3D structure models for target protein sequences of unknown structure. In particular, fold recognition or threading is aimed at those targets sharing little or no significant sequence similarity to any protein of known structure (Fischer et al., 1996). The LiveBench continuous benchmarking program assesses the performance of automatic fold-recognition servers by submitting as prediction targets the sequences of newly released PDB entries with no clear sequence similarity to previously released proteins. After collecting the predicted models from the participating servers, the predictions are assessed by comparing them to the experimental structures. Among the target sequences recently considered, were a number of chains of the newly released PDB entry 1kc9, which describes the crystal structure of the large ribosomal subunit from the bacterium D. radiodurans (Harms, 2001). This is a Cα-only structure with a resolution of 3.1 A. Here we focus on the M chain, which corresponds to the structure of residues 2-114 of the ribosomal protein L18, herein referred to as 1kc9 M. All the fold-recognition methods (Bujnicki et al., 2001b) participating in LiveBench (including iterated PSI-BLAST searches) suggested with very high confidence that a