Intruding into the midnight zone of protein comparisons

描述(申请人提供)：蛋白质序列的“黄昏地带” 比较是指序列相似性不足以满足以下条件的区域 得出结构相似的结论。在所有蛋白质对中的绝大多数 相似的结构存在于‘午夜地带’，即它们的序列也不同 很大程度上是基于序列的比较。在这里，我们建议细化、扩展和 结合序列比对、结构预测和 功能信息。目标是揭开隐藏的相似之处 用一种可靠的自动化工具对生物体进行排序。接近尾声时，我们的 项目中，生命实现的大多数蛋白质家族的序列将 据说是有空的。我们希望我们的系统将正确地检测到 这些中的大多数都有关系。 (1)基于预测的线程将序列比对与预测相结合 二级结构和可访问性，以找到遥远的相似之处。我们希望 通过比较族显著提高检测和对齐精度 具有家族而不是单一蛋白质。(2)约三分之一的蛋白质 在蠕虫和苍蝇中，似乎有较长的区域，缺乏规则的二级结构。 我们希望开发一种方法来可靠地检测和比较这种情况 地区。(3)目前还没有一种方法能在极端不同的 膜蛋白。我们建议开发一种结合膜的这种方法 通过对膜蛋白的分类进行处理。(4)开始顺序 在黄昏区域和更低的区域进行比较是一项极其艰巨的任务，大多数 现有的方法准确度很低。在实践中，专家们将 正在研究的蛋白质对之间的功能方面。我们想要 开发一种自动评估功能方面的方法。特别是，我们 打算从与DNA结合的蛋白质开始。任务将是(I)预测 蛋白质中的DNA结合位点，以及(Ii)限制对亚集的穿线 找到了结合区的蛋白质。在接下来的步骤中，我们 希望利用通用序列基序进行自动比对。(5)穿线 整个基因组：第一个任务将是找到整个基因组中的所有蛋白质 我们知道其结构的有机体。然而，我们的特殊优势是 方法将是寻找遥远的相似之处，即使在没有实验的情况下 有关结构的信息。

Description (provided by applicant): The 'twilight zone' of protein sequence comparison is the region in which sequence similarity does not suffice to conclude e.g. structural similarity. The vast majority of all protein pairs of similar structure populate a 'midnight zone' i.e. their sequences differ too much for sequence-based comparisons. Here, we propose to refine, extend, and specialise methods combining sequence alignment, structure prediction and functional information. Goal is to unravel hidden similarities in entirely sequenced organisms by a reliable, automatic tool. Towards the end of our project, the sequences for most protein families realised by life will supposedly be available. We hope that our system will correctly detect a relation for most of these. (1) Prediction-based threading combines sequence alignments with predictions of secondary structure and accessibility to find remote similarities. We hope to considerably improve detection and alignment accuracy by comparing families with families rather than single proteins. (2) About one third of all proteins in worm and fly seem to have long regions lacking regular secondary structure. We hope to develop a method tailored to reliably detect and compare such regions. (3) No current method finds similarities between extremely diverged membrane proteins. We propose to develop such a method combining 'membrane threading' with classifications of membrane proteins. (4) Since sequence comparison in the twilight zone and below is an extremely demanding task, most existing methods have very low levels of accuracy. In practice, experts compare aspects of function between the protein pair under investigation. We want to develop an automatic method evaluating functional aspects. In particular, we intend to start with proteins binding to DNA. The tasks will be to (i) predict DNA-binding sites in proteins, and to (ii) restrict the threading to the subset of proteins for which binding regions were found. In the following step, we hope to use general sequence motifs for the automatic comparison. (5) Threading entire genomes: the first task will be to find all proteins in an entire organism for which we know structure. However, the particular edge of our method will be to find remote similarities even in the absence of experimental information about structure.