BIOINFORMATICS, DISORDERED PROTEINS, AND FUNCTION

生物信息学的一个主要目标是从序列中预测蛋白质的结构/功能；目前最成功的方法是序列同源性。目前这些基于同源的方法显式或隐式地假设范式序列产生结构产生函数。由于蛋白质的功能取决于灵活性、运动，甚至在某些情况下缺乏结构（紊乱），将各种运动预测与结构预测结合起来应该可以改善功能预测。从序列和局部无序预测灵活性的算法已经开发出来，正如预测在两种结构状态之间切换的序列的算法一样。序列复杂度也可能是迁移率的一种间接度量，还可以通过研究已知的例子来发现其他度量。我们建议确定同源性预测的相互作用，这些预测仅使用序列信息，并明确表示结构信息和从氨基酸序列确定的运动信息。结构信息将由螺旋、薄片等、预测和疏水力矩计算来表示。运动信息将由灵活性预测、开关序列预测、有序/无序预测、序列复杂性以及在此工作过程中发现的新措施来表示。由于不同的函数会在不同程度上涉及到运动参数，因此计划是在序列族的基础上应用这种组合方法。提出了一种新的比较方法，称为属性轮廓，用于表示结构和运动信息。Gribskov/Eisenberg 1D剖面和相关的属性剖面将使用神经网络数据模型组合成单个预测。预测结果将与实验观测结果进行比较，并使用折刀法进行评估。如果成功，这项工作将改善氨基酸序列对蛋白质功能的预测，这对于各种基因组计划中大量功能未知的蛋白质序列是很重要的。运动和无序是需要添加到蛋白质结构/功能表征的重要部分。例如，紫杉醇最近被证明与Bcl-2中的一个无序环结合。阿尔茨海默病、传染性海绵状脑病、帕金森氏病以及诸如金黄色葡萄球菌、口蹄疫病毒和（可能）艾滋病毒等传染性病原体严重依赖于蛋白质紊乱区域。因此，了解序列、灵活性、有序/无序、复杂性、结构和功能之间的相互作用显然与人类健康有关。

A major objective of bioinformatics is to predict protein structure/function from sequence; the most successful current methods use sequence homology. These current homology-based methods explicitly or implicitly assume the paradigm sequence yields structure yields function. Since protein function depends on flexibility, movement, or even lack of structure (disorder) in some cases, combining various motion predictions with structure predictions should improve prediction of function. Algorithms for predicting flexibility from sequence and local disorder have been developed, as have algorithms to predict sequences that snitch between two structured states. Sequence complexity might also be an indirect measure of mobility and still other measures could be discovered by the study of known examples. We propose to determine the interplay of homology-based predictions that use sequence information only with explicit representations of both structure information and motion information as determined from amino acid sequence. Structural information will be represented by helix, sheet, etc., predictions, and by hydrophobic moment calculations. Motion information will be represented by flexibility prediction, switch sequence prediction, order / disorder prediction, sequence complexity, and new measures if any are discovered in the course of this work. Since different functions would involve motional parameters to different extents, the plan is to apply this combined approach on a sequence family basis. Novel comparisons, called Attribute Profiles, are proposed for the representation of structure and motion information. Sets of Gribskov/Eisenberg 1D Profiles and associated Attribute Profiles will be combined into single predictions using neural network data models. Prediction outcomes will be compared with experimental observations and evaluated using the jackknife method. If successful, this work will improve prediction of protein function from amino acid sequence, which is important given the significant numbers of protein sequences with undetermined functions coming out of the various genome projects. Motion and disorder are important pieces that need to be added to the characterization of protein structure/function. For example, taxol has been shown recently to bind to a disordered loop in Bcl-2. Alzheimer disease, transmissible spongiform encephalopathies, Parkinson disease and infectious agents such as Staphylococcal aureus, foot-and-mouth disease virus, and (perhaps) HIV depend critically on disordered regions of protein. Thus, understanding the interplay of sequence, flexibility, order / disorder, complexity, structure and function clearly relates to human health.