Viksna Juris, Gilbert David
Institute of Mathematics and Computer Science, University of Latvia, Rainis boulevard 29, Riga LV-1459, Latvia.
Bioinformatics. 2007 Apr 1;23(7):832-41. doi: 10.1093/bioinformatics/btm022. Epub 2007 Feb 4.
The evolution of protein sequences can be described by a stepwise process, where each step involves changes of a few amino acids. In a similar manner, the evolution of protein folds can be at least partially described by an analogous process, where each step involves comparatively simple changes affecting few secondary structure elements. A number of such evolution steps, justified by biologically confirmed examples, have previously been proposed by other researchers. However, unlike the situation with sequences, as far as we know there have been no attempts to estimate the comparative probabilities for different kinds of such structural changes.
We have tried to assess the comparative probabilities for a number of known structural changes, and to relate the probabilities of such changes with the distance between protein sequences. We have formalized these structural changes using a topological representation of structures (TOPS), and have developed an algorithm for measuring structural distances that involve few evolutionary steps. The probabilities of structural changes then were estimated on the basis of all-against-all comparisons of the sequence and structure of protein domains from the CATH-95 representative set. The results obtained are reasonably consistent for a number of different data subsets and permit the identification of several 'most popular' types of evolutionary changes in protein structure. The results also suggest that alterations in protein structure are more likely to occur when the sequence similarity is >10% (the average similarity being approximately 6% for the data sets employed in this study), and that the distribution of probabilities of structural changes is fairly uniform within the interval of 15-50% sequence similarity.
The algorithms have been implemented on the Windows operating system in C++ and using the Borland Visual Component Library. The source code is available on request from the first author. The data sets used for this study (representative sets of protein domains, matrices of sequence similarities and structural distances) are available on http://bioinf.mii.lu.lv/epsrc_project/struct_ev.html.
蛋白质序列的进化可以通过一个逐步的过程来描述，其中每一步都涉及少数氨基酸的变化。类似地，蛋白质折叠的进化至少可以部分地通过一个类似的过程来描述，其中每一步都涉及影响少数二级结构元件的相对简单的变化。其他研究人员此前已经提出了许多经生物学证实的此类进化步骤的例子。然而，与序列的情况不同，据我们所知，尚未有人尝试估计不同类型此类结构变化的相对概率。
我们试图评估一些已知结构变化的相对概率，并将这些变化的概率与蛋白质序列之间的距离联系起来。我们使用结构的拓扑表示（TOPS）对这些结构变化进行了形式化，并开发了一种用于测量涉及少数进化步骤的结构距离的算法。然后，基于来自CATH - 95代表性集的蛋白质结构域的序列和结构的全对全比较，估计了结构变化的概率。对于许多不同的数据子集，所获得的结果相当一致，并允许识别蛋白质结构中几种“最常见”的进化变化类型。结果还表明，当序列相似性>10%时（本研究中使用的数据集的平均相似性约为6%），蛋白质结构的改变更有可能发生，并且在15 - 50%的序列相似性区间内，结构变化概率的分布相当均匀。
这些算法已在Windows操作系统上用C++并使用Borland可视化组件库实现。可向第一作者索取源代码。本研究使用的数据集（蛋白质结构域代表性集、序列相似性矩阵和结构距离矩阵）可在http://bioinf.mii.lu.lv/epsrc_project/struct_ev.html上获取。
