Merizo: a rapid and accurate domain segmentation method using invariant point attention
Merizo: a rapid and accurate domain segmentation method using invariant point attention
复制标题
Merizo：一种使用不变点注意力的快速准确的域分割方法
DOI：
10.1101/2023.02.19.529114
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Lau A
中科院分区：
文献类型：
--
作者：
Lau A
蛋白质结构域是蛋白质结构中独特的、模块化的、局部紧凑的单元，它可以独立于蛋白质的其他部分折叠和发挥功能。识别对应于结构域的区域是不平凡的，并且结构域折叠的分类已经在诸如CATH和ECOD的数据库中被广泛记录。随着AlphaFold2生成的2亿个蛋白质模型的出现，将蛋白质准确分解为其组成结构域的能力将允许深入研究其组成，并实现许多新的研究方向。虽然有许多现有的方法来识别蛋白质中的结构域，但它们要么不准确，要么运行效率太高，要么不能处理不连续的结构域。在这里，我们描述了我们基于深度学习的域分割方法Merizo，它与其他传统方法有很大的不同，通过学习直接将残基聚类到域中，以自下而上的方式进行分割。我们的网络在CATH域上进行了完全端到端的训练，并且在预测正确的边界位置以及匹配整体域拓扑结构方面都优于当前最先进的方法。Merizo将在https://github.com/psipred/Merizo上提供。
AbstractProtein domains are distinct, modular and locally compact units of protein structures which may fold and function independently to the rest of the protein. Identifying the regions corresponding to a domain is non-trivial, and the classification of domain folds have been extensively documented in databases such as CATH and ECOD. With the advent of 200 million protein models generated by AlphaFold2, the ability to accurately decompose proteins into their constituent domains will allow a deep dive into their compositions and enable many new lines of research. Although there are many existing methods for identifying domains in proteins, they are either inaccurate, too efficient to run, or do not handle discontinuous domains. Here, we describe our deep learning-based approach for domain segmentation called Merizo, which differs significantly from other conventional methods by conducting segmentation in a bottom-up manner by learning to directly cluster residues into domains. Our network is trained fully end-to-end on CATH domains and outperforms current state-of-the-art methods both in predicting correct boundary positions as well as matching the overall domain topology. Merizo will be made available at https://github.com/psipred/Merizo.