Integrated in silico prediction of protein interaction motifs using interactome networks and high-resolution 3-dimensional structures

该项目旨在确定蛋白质上对它们与其他蛋白质相互作用至关重要的位置。许多蛋白质之间的相互作用是由短线性基序(SLIMs)介导的：蛋白质的短伸展(5-15个氨基酸长)，其中只有几个位置对功能至关重要。这些基序对于至关重要的生物过程是至关重要的，例如信号通路和将蛋白质定位到细胞的正确部分。这个项目的主要目标是集成一些领先的计算技术来预测新的苗条，并在这样做的同时，为蛋白质-蛋白质相互作用网络增加关键的细节。这将产生一个有价值的潜在瘦身资源，包括确定的事件和相互作用。鉴于SLIM介导的相互作用作为未来治疗靶点的巨大前景，这种资源可能成为制药业和未来药物设计的潜在金矿。任何预测性生物信息学分析的一个关键部分是重新发现以前已知的结果。真核生物线性基序(ELM)数据库提供了丰富的已知真核基序资源，这些资源将形成对在调查过程中(重新)发现的已知瘦身进行注释的基础。该项目期间产生的数据将反过来反馈给ELM资源，以改进现有的ELM注释，并可能提供已知主题的新注释出现。除了有助于进一步研究特定的蛋白质-蛋白质相互作用，这些预测/注释也将引起更广泛的科学界的兴趣，他们有兴趣了解蛋白质相互作用的基本原理。为了生成SLIM预测，该项目将为两个不同但相关的活动整合尖端工具：(1)从蛋白质的3D结构预测参与SLIM介导的相互作用的蛋白质区域；(2)从所有共享共同相互作用伙伴的蛋白质中识别过度代表的重复序列模式。首先，将利用结构特征来确定特定蛋白质中的候选区域。Slims通常与其伴侣蛋白质中较大的球状结构域相互作用，这种“结构域-基序相互作用”的结构特征可以用来突出可能的基序区域。然后将这些区域与已知与候选蛋白质相互作用的其他蛋白质进行比较。目前，最成功的方法是显式地使用收敛进化模型进行检测。在该模型下，识别的共同基序必须由没有其他可检测到的序列相似性的序列共享。此前，我们在基准数据上开发了这些工具中最成功的SLiMFinder，它既考虑了输入蛋白质之间的进化关系，也考虑了正在搜索的总基序空间，以估计过度代表基序的统计意义。对于这个项目，将使用SLiMFinder的扩展，利用3D方法将在其中一个蛋白质上识别特定的短区域这一事实。这些额外的信息使该方法变得更加敏感。这些结果将对任何试图了解蛋白质-蛋白质相互作用和信号通路的分子基础的人非常感兴趣。在项目过程中，将进一步开发和验证所使用的方法，为今后的调查提供有用的工具。将提供开放源码软件和网络服务器实施，以促进进一步的应用。这将使这些方法可供研究特定蛋白质和相互作用的普通科学家使用。

This project aims to identify sites on proteins that are critical for their interactions with other proteins. Many protein-protein interactions are mediated by Short Linear Motifs (SLiMs): short stretches of proteins (5-15 amino acids long), of which only a few positions are critical to function. These motifs are vital for biological processes of fundamental importance, such as signalling pathways and targeting proteins to the correct part of a cell. The primary objective of this project is to integrate a number of leading computational techniques to predict novel SLiMs and, in so doing, add crucial detail to protein-protein interaction networks. This will generate a valuable resource of potential SLiMs, including defined occurrences and interactions. Given the great promise SLiM-mediated interactions hold as future therapeutic targets, this resource could be a potential gold-mine for the pharma industry and future drug design. A crucial part of any predictive bioinformatics analysis is the rediscovery of previously known results. The Eukaryotic Linear Motif (ELM) database provides a rich resource of known eukaryotic motifs and these will form the basis of annotating known SLiMs that are (re)discovered during the course of the investigation. Data generated during this project will, in turn, be fed back into the ELM resource to improve existing ELM annotation and, potentially, provide new annotated occurrences of known motifs. As well as being useful for further investigation of specific protein-protein interactions, these predictions/annotations will be of interest to the wider scientific community who are interested in understanding the fundamental principles of how proteins interact with each other. To generate the SLiM predictions, this project will put together cutting-edge tools for two distinct but related activities: (1) predicting regions of proteins involved in SLiM-mediated interactions from 3D structures of proteins; (2) identifying over-represented recurring sequence patterns from proteins that all share a common interaction partner. First, structural features will be used to identify candidate regions in specific proteins. SLiMs typically interact with larger, globular domains in their partner protein, and structural signatures of such 'domain-motif interactions' can be used to highlight possible motif regions. These regions will then be compared to other proteins known to interact with the same protein as the candidate. Currently, the most successful approaches for this explicitly use a model of convergent evolution for detection. Under this model, the common motif identified must be shared by sequences that have no other detectable sequence similarity. Previously, we developed the most successful of these tools on benchmarking data, SLiMFinder, which accounts for both the evolutionary relationships found between input proteins and the total motif space being searched to estimate the statistical significance of over-represented motifs. For this project, an extension of SLiMFinder will be used that takes advantage of the fact that the 3D methods will have identified a specific short region on one of the proteins. This extra information makes the method much more sensitive. These results will be of great interest to anyone trying to understand the molecular basis of protein-protein interactions and signalling pathways. During the course of the project, the methods used will be further developed and validated, providing useful tools for future investigations. Open source software and webserver implementations will be made available to facilitate further application. This will make these methods available to bench scientists studying specific proteins and interactions.