Development and assessment of methods for membrane protein structure prediction

在进化过程中，蛋白质保留了共同的三维结构特征，尽管潜在的氨基酸序列可能会有很大的差异。这种关系甚至可以在给定的蛋白质结构中找到，这种关系是由定义的元件的复制和重复引起的。因此，识别两个蛋白质或同一蛋白质的两个区域之间的关系是非常有价值的，因为它们是非常遥远的亲缘关系。大多数情况下，需要在初级氨基酸编码的水平上识别这些关系，这是通过比对它们的序列来实现的。然而，当确定了结构后，可以通过覆盖公共区域来检测这种关系，这是一种称为结构对齐的技术。这两个过程都涉及到相当大的挑战，特别是在两种蛋白质之间的相似性很小的情况下。因此，仍然需要可靠和准确地计算序列或结构比对的方法。在过去的一年里，我们从两个不同的方面共同努力开发了这些工具。 首先，我们对我们的同源膜蛋白结构的基准集进行了改进，其早期版本被称为HOMP。用来编译数据集的代码现在用的是Python编程语言，更精简，可以并行运行，随着可用膜蛋白结构数据库继续假指数增长，允许未来快速更新。这些变化将促进我们之前开发的序列比对软件AlignMe的再培训，从而改进我们的软件。 其次，我们对膜蛋白结构对称性的早期(人工)分析进行了扩展。我们首先评估了现有对称性检测方法的可靠性，开发了一组包含对称性的蛋白质(称为MemSTATS)，并测试了现有方法在检测这些对称性方面的准确性。然后，我们基于两个可用的对称性分析工具(SymD和CEsymm)开发了一个协议，并将其系统地应用于已知的蛋白质结构。这种方法有望识别对称和不对称膜蛋白的模式和关系，并揭示这些对称性如何与功能机制相关。对称性数据集和所有相关代码已通过在线数据储存库提供。 这两种分析现在已经被合并到一个名为Enneass(按结构和对称性分析的膜蛋白百科全书)的数据库中。我们最近已采取步骤，整合作为数据库两个组成部分基础的软件。这一组合使我们能够利用有关结构邻居的信息来提高对称性分析的质量和适用性。此外，在NINDS内部生物信息学工作人员(Yavaktar和Kumar)的帮助下，我们通过设在https://encompass.ninds.nih.gov.的公共网络服务器完全实现了数据的可视化在过去的一年里，我们大大扩展了数据库的搜索功能，并改进了网络界面的可用性和功能。参见参考文献1. 在这些研究的同时，我们还展示了使用NHBLI在Faraldo-Gomez实验室开发的名为EBmetaD的先进分子模拟方法，以提供电子顺磁共振距离分布的定量解释。我们将这种方法应用于我们的合作者(Ziegler，Regensburg和Prisner，法兰克福)在模拟甜菜碱转运体的已知结构时获得的数据，这使我们能够区分在生理相关条件下可能出现在实验集合中的状态；参考文献。2.这种方法对于确定神经元转运蛋白如SERT和DAT的构象周期中存在的状态是有用的。

During evolution, proteins retain common three-dimensional structural features, even though the underlying sequence of amino acids can diverge dramatically. Such relationships can even be found internally within a given protein structure, arising from duplication and repetition of defined elements. Identifying relationships between two proteins, or two regions of the same protein, that are very distantly related, therefore, can be extremely valuable. Most often, identification of those relationships is needed on the level of primary amino acid codes, which is achieved by aligning their sequences. However, when structures have been determined, such relationships can be detected by overlaying common regions, a technique known as structure alignment. Both procedures involve considerable challenges, especially when the similarities between the two proteins are small. Consequently, there remains a need for methods that reliably and accurately compute sequence or structure alignments. In the past year, we have combined efforts from two different fronts in developing such tools. First, we have made improvements to our benchmark set of homologous membrane protein structures, earlier versions of which were called HOMEP. The code used to compile the dataset is now in the Python programming language, is more streamlined and can run in parallel, allowing for fast future updates as the database of available membrane protein structures continues its pseudo-exponential growth. These changes will facilitate retraining of, and therefore improvements in, our sequence alignment software, AlignMe developed previously. Second, we have expanded upon an earlier (manual) analysis of symmetries in structures of membrane proteins. We first assessed the reliability of available symmetry-detection approaches, by developing a benchmark set of symmetry-containing proteins (called MemSTATS), and testing available methods for their accuracy at detecting these symmetries. We then developed a protocol based on two of these available symmetry analysis tools (SymD and CEsymm), and apply it systematically to known protein structures. This approach is expected to identify patterns and relationships in symmetrical and asymmetrical membrane proteins and to reveal how those symmetries relate to functional mechanisms. The symmetry datasets and all relevant code have been made available through an online data repository. These two analyses have now been combined into a single database called EncoMPASS (Encyclopedia for Membrane Proteins Analyzed by Structure and Symmetry). We have recently made steps to integrate the software underlying the two components of the database. The combination allows us to leverage information about structural neighbors in order to improve the quality and applicability of the symmetry analysis. Moreover, with the aid of the NINDS intramural Bioinformatics staff (Yavaktar and Kumar), we have made visualization of the data fully accessible through a public webserver hosted at https://encompass.ninds.nih.gov. In the past year, we have substantially expanded the search functions for the database, as well as improving the usability and functionality of the web interface. See Ref. 1. In parallel to these studies, we have demonstrated the use of an advanced molecular simulation methodology called EBmetaD developed at NHBLI in the Faraldo-Gomez lab, for providing a quantitative interpretation of electron paramagnetic resonance distance distributions. We applied this method to data obtained by our collaborators (Ziegler, Regensburg, and Prisner, Frankfurt) in simulations of known structures of a betaine transporter, which enabled us to distinguish between states that are likely to be present in the experimental ensemble under physiologically-relevant conditions; ref. 2. This method should be useful for defining the states present in conformational cycles of neuronal transporters such as SERT and DAT.