Center for Critical Assessment of Structure Prediction (CASP)

项目总结 蛋白质结构的实验测定通常提供原子精度模型，但本质上是时间- 消耗，往往昂贵，而且并不总是可能的。计算建模目前的精确度较低，但提供了 当实验结果不可用时，另一种方法。CASP(结构临界评估)的目标 预测)是通过进行社区范围的客观实验来推进蛋白质结构建模领域 确定当前方法的优点和缺点，从而促进进步。大约100个研究小组 全球范围的参与。在最近的一次实验(2018)中，有9个建模类别的5.7万份提交， 包括三万五千多个三级结构模型。该中心为CASP提供基础设施，AIM 1是继续 这一资源的开发和运营。主要任务包括登记和与参与者沟通； 模型目标的征集、表征和管理；模型的收集和验证；以及 意见书的数字分析。这些操作得到安全可靠的数据基础设施的支持。《中心》 还开发评估、分析和显示软件，并提供对模型和评估结果的访问。CASP依赖 对独立评价者、建模专家或相关实验领域的专家，对结果进行解读。《中心》 协调这一进程，提供评价数据，并在必要时实施新的评价方法。 最近的CASP表明，模型精度有了显著的提高，特别是对于最困难的情况 不能使用同源建模。推动进步的一个主要因素是使用新的机器学习方法， 尤其是卷积神经网络。这些方法和相关方法似乎准备在以下方面取得进一步的重大进展 一些关键的建模领域。该项目下一阶段的计划旨在利用这些和其他 进步的机会。在对小蛋白和结构域进行建模方面取得的更大成功要求将重点转移到 仍然具有挑战性的大型多域蛋白质和复合体领域(目标2)，预计这方面的进展既来自 机器学习的发展和稀疏实验数据的结合。虽然模型的准确性 虽然已经有所改善，但仍然很少与实验竞争。目标3是追求使模特更多的战略 准确和有用，通过在改进初始模型方面取得进一步进展，更好地估计模型精度， 以及对模型的实用性进行评估。目标4介绍了加强CASP与 更广泛的研究社区，提供直接解决当代问题的模型(例如，针对CoV-2 蛋白质结构)，并通过会议、网络研讨会、出版物和其他方式促进交流。

PROJECT SUMMARY Experimental determination of protein structure often provides atomic accuracy models, but is inherently time- consuming, often costly, and not always possible. Computational modeling is currently less accurate, but offers an alternative approach when experimental results are not available. The goal of CASP (Critical Assessment of Structure Prediction) is to advance the protein structure modeling field by conducting community-wide experiments that objectively determine the strengths and weaknesses of current methods and so foster progress. Approximately 100 research groups world-wide participate. In the most recent experiment (2018), there were 57,000 submissions in nine modeling categories, including over 35,000 tertiary structure models. The Center provides the infrastructure for CASP and Aim 1 is the continued development and operation of this resource. Principal tasks include registration and communication with participants; solicitation, characterization, and management of modeling targets; collection and validation of models; and extensive numerical analysis of submissions. These operations are supported by a secure and robust data infrastructure. The Center also develops evaluation, analysis, and display software, and provides access to models and evaluation results. CASP relies on independent assessors, experts in modeling or a related experimental field, to interpret the results. The Center coordinates this process, providing evaluation data and, when necessary, implementing new evaluation methods. Recent CASPs have shown dramatic improvements in model accuracy, especially for the most difficult cases where homology modeling cannot be used. A major factor driving progress is the use of new machine learning approaches, particularly convolutional neural networks. These and related methods appear poised to make further major advances in a number of key modeling areas. The plan for the next period of the project is designed to capitalize on these and other opportunities for progress. Greater success with modeling small proteins and domains dictates a shift in emphasis to the still challenging area of large multi-domain proteins and complexes (Aim 2), where progress is expected both from the machine learning developments and from the incorporation of sparse experimental data. Although accuracy of models has improved, it is still seldom competitive with experiment. Aim 3 is to pursue strategies that will make models more accurate and useful, by nurturing further progress in refining initial models, better methods for estimating model accuracy, and assessment of the utility of models. Aim 4 introduces new ways of strengthening interactions between CASP and the broader research community, providing models that directly address contemporary problems (for example, for CoV-2 protein structures) and boosting communications through meetings, webinars, publications and other means.