REFINEMENT OF PREDICTED LOW-RESOLUTION PROTEIN MODELS TO HIGH-RESOLUTION ALL-AT

将预测的低分辨率蛋白质模型细化为高分辨率 All-AT

• 批准号: 7723173
• 负责人: JEFFREY SKOLNICK
• 金额: $ 0.05万
• 依托单位: CARNEGIE-MELLON UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2008
• 资助国家: 美国
• 起止时间: 2008-08-01 至 2009-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/7723173
• 关键词: Affect; Algorithms; Amino Acid Sequence; Amino Acids; Computer Retrieval of Information on Scientific Projects Database; Data; Distant; Drug Design; Environment; Failure; Free Energy; Funding; Generations; Goals; Grant; Institution; Laboratories; Ligands; Methodology; Modeling; Molecular Conformation; Pharmaceutical Preparations; Plant Roots; Proteins; Relative (related person); Research; Research Personnel; Resolution; Resources; Screening procedure; Shapes; Solvents; Source; Structure; United States National Institutes of Health; Vertebral column; improved; molecular mechanics; protein structure; protein structure prediction

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. A major challenge in protein structure prediction is the refinement of low-resolution predicted models (e.g. those with a backbone root-mean-square-deviation, RMSD, from the native structure of about 5 ), to high-resolution all-atom structures with a RMSD of less than 2 . To date, there have only been a few documented instances where structures at atomic detail have improved relative to their initial starting conformation. Even if low-resolution models with a RMSD from the native structure of about 5 were available for a target protein, the refinement to a high-resolution structure with a RMSD of less than 2 can be difficult to achieve. Given the increasing ability of protein structure prediction algorithms, such as TASSER developed in our laboratory, to assemble approximately correct native structures, the effort should be made to refine such models so that they can be used for ligand screening as well as more general types of functional annotation. High-resolution structures are useful in detailed studies of mechanisms as well as in the design of drug molecules in which the (atomic) local environment of the small (drug) molecule requires accurate characterization. While significant progress has been made in the generation of low-resolution structures and their selection, the final step of structural adjustment still poses significant difficulties. Despite considerable effort, structure refinement from proteins having near native structures has proven to be very difficult, especially if all atom models and atomic potentials are used. It is necessary for successful refinement of low-resolution models that the force field recognizes the native structure of a protein as the lowest free energy minimum, and it shows correlation between energy and the native-likeness of decoy structures. Current all-atom force fields in many cases fail to recognize the native structure as the lowest free energy minimum and energy of the decoys do not monotonically decreases as the native conformation is approached. The source of this inaccuracy may be: a). the mathematical form of the potential b). lack of important physical interactions in the potential, e.g. polarization or correlation effects (many body interactions), or c). inadequate parameterization of the force field, which does not include sufficient information about the global shape of the energy landscape. Most all-atom force fields rely on the assumption that the conformation of each residue is energetically insensitive to the conformation of its neighbors (i.e. they do not contain explicit correlation terms for the backbone conformation (torsional angles) of the neighboring residues). However, there are experimental and statistical data from solved protein structures, showing that the backbone conformation of a residue in the amino acid chain is correlated with the conformation and identity of the neighboring residues. These data reflect the influence of the environment of the protein, where the correlations of local effects with solvent and non-local effects are non-separable. Therefore, it is important to determine to what extent the local interactions (between residues close in amino acid sequence) alter the accessible conformational space and how much of this effect originates from non-local interactions (between residues close in space but distant in sequence) and solvation, and whether current extant force fields, that do not explicitly include backbone correlation and polarization can reproduce these effects. Conceivably, ignoring some important interactions could result in inaccuracies of the relative conformational energies thereby affecting the ability to identify the native structure as the lowest energy minimum. However, if the functional form of current molecular mechanics potentials is mathematically correct, then the failure to recognize the native structure as the global energy minimum could arise from incorrect details of the force field parameters that should be fixable by reparameterization. The long-term goal of this project is to evaluate and improve accuracy of current all-atom force fields and develop methodology to refine the low-resolution protein models, resulted from the protein structure prediction approaches developed by Skolnick and coworkers, to high-resolution structures in all-atom representation.

这个子项目是许多研究子项目中利用 资源由NIH/NCRR资助的中心拨款提供。子项目和 调查员(PI)可能从NIH的另一个来源获得了主要资金， 并因此可以在其他清晰的条目中表示。列出的机构是 该中心不一定是调查人员的机构。 蛋白质结构预测的一个主要挑战是将低分辨率预测模型(例如，那些主干均方根偏差(RMSD)约为5的模型)改进为RMSD小于2的高分辨率全原子结构。到目前为止，只有少数几个有记录的例子表明，原子细节的结构相对于它们最初的起始构象有所改善。即使具有来自天然结构的RMSD约为5的低分辨率模型可用于目标蛋白质，精细化到RMSD小于2的高分辨率结构也可能很难实现。鉴于蛋白质结构预测算法，如我们实验室开发的Tasser，组装近似正确的天然结构的能力越来越强，因此应该努力改进这样的模型，以便它们可以用于配体筛选以及更一般类型的功能注释。高分辨结构在机理的详细研究以及药物分子的(原子)局部环境需要精确表征的药物分子的设计中都是有用的。虽然在低分辨率结构的产生和选择方面取得了重大进展，但结构调整的最后一步仍然构成重大困难。尽管付出了相当大的努力，但事实证明，从具有接近天然结构的蛋白质中精炼结构是非常困难的，特别是如果使用所有的原子模型和原子势。对于低分辨率模型的成功改进，力场必须将蛋白质的天然结构识别为最低自由能最小值，并显示能量与诱饵结构的天然相似性之间的相关性。目前的全原子力场在许多情况下都不能识别本征结构为最低自由能最小值，并且随着本征构象的接近，诱骗分子的能量并不单调地减小。这种不准确的来源可能是：a)。势b)的数学形式。在潜势中缺乏重要的物理相互作用，例如极化或相关效应(许多身体相互作用)，或c)。力场的参数化不充分，没有包括关于能源格局全球形状的充分信息。大多数全原子力场依赖于这样的假设，即每个残基的构象在能量上对其邻近残基的构象不敏感(即它们不包含关于相邻残基的主干构象(扭转角)的显式关联项)。然而，从已求解的蛋白质结构中获得的实验和统计数据表明，氨基酸链中一个残基的主干构象与相邻残基的构象和同一性有关。这些数据反映了蛋白质环境的影响，其中局部效应与溶剂和非局部效应的相关性是不可分割的。因此，重要的是确定局部相互作用(氨基酸序列中接近的残基之间)在多大程度上改变可访问的构象空间，这种影响在多大程度上源于非局部相互作用(空间接近但序列距离较远的残基之间)和溶剂化，以及当前不明确包括主链相关和极化的现有力场是否能够复制这些影响。可以想象，忽略一些重要的相互作用可能会导致相对构象能量的不准确，从而影响将自然结构识别为最低能量最小值的能力。然而，如果当前分子力学势的泛函形式在数学上是正确的，那么未能将自然结构识别为全局能量最小值可能是因为力场参数的不正确细节，而力场参数应该可以通过重新参数化来固定。该项目的长期目标是评估和提高当前全原子力场的准确性，并开发方法学，将由Skolnick和他的同事开发的蛋白质结构预测方法产生的低分辨率蛋白质模型改进为全原子表示的高分辨率结构。