Integrated Prediction of Protein Struture at 1D, 2D and 3D Levels

描述（由申请人提供）：从氨基酸序列计算预测蛋白质结构是生物信息学和计算生物学中最重要和最具挑战性的问题之一。随着后基因组时代蛋白质序列的指数级增长，人们迫切需要精确的蛋白质结构预测方法和工具。在这里，我们建议开发一个集成的方法来推进蛋白质结构预测在1维（1D），2维（2D）和3维（3D）水平。在1D水平上，新的信息，如结构域进化信号，选择性基因剪接位点，和2D蛋白质接触图将被用来预测蛋白质结构域边界的序列。在2D水平上，残基接触传播、机器学习增强、线性规划和马尔可夫链蒙特卡罗模拟等新方法将用于推进结构域或蛋白质的残基-残基接触预测。在3D层面，将使用2D接触预测、通过机器学习的折叠识别和多模板组合来增强基于模板和从头算结构预测。最后，将开发基于知识的统计机器学习方法和模型组合算法，以可靠地评估和改进预测蛋白质结构模型的质量。这种方法的几个创新方面之一是整合1D，2D和3D预测，以便通过蛋白质结构单元-结构域相互改进。1D，2D和3D蛋白质结构预测方法将作为用户友好的软件包和Web服务发布给科学界。这些工具和网络服务将是有用的蛋白质结构预测，结构测定，功能分析，蛋白质工程，蛋白质诱变分析和蛋白质设计。 公共卫生相关性：该项目将为基础生物医学研究开发精确的计算方法和工具，如蛋白质结构预测，蛋白质功能分析，蛋白质设计，蛋白质工程和基于结构的药物设计。

DESCRIPTION (provided by applicant): Computational prediction of protein structure from the amino acid sequence is one of the most important and challenging problems in bioinformatics and computational biology. With the exponential growth of protein sequences without solved protein structures in the post-genomic era, accurate protein structure prediction methods and tools are in urgent need. Here, we propose to develop an integrated approach to advance protein structure prediction at the 1-dimensional (1D), 2-dimensional (2D) and 3-dimensional (3D) levels. At the 1D level, novel information such as domain evolution signals, alternative gene splicing sites, and 2D protein contact map will be used to predict protein domain boundaries from the sequences. At the 2D level, new methods such as residue contact propagation, machine learning boosting, linear programming, and Markov Chain Monte Carlo simulations will be used to advance residue-residue contact prediction for a domain, or a protein. At the 3D level, 2D contact prediction, fold recognition via machine learning, and multi-template combination will be used to enhance both template-based and ab initio structure prediction. Finally, knowledge-based statistical machine learning methods and model combination algorithms will be developed to reliably evaluate and refine the quality of predicted protein structural models. One of several innovative aspects of this approach is to integrate 1D, 2D, and 3D predictions in order to improve each other through protein structural unit - domains. The 1D, 2D, and 3D protein structure prediction methods will be implemented as user-friendly software packages and web services released to the scientific community. These tools and web services will be useful for protein structure prediction, structure determination, functional analysis, protein engineering, protein mutagenesis analysis, and protein design. PUBLIC HEALTH RELEVANCE: The project will develop accurate computational methods and tools for basic biomedical research such as protein structure prediction, protein function analysis, protein design, protein engineering, and structure-based drug design.