When does structure modeling go wrong? A PDB-scale analysis of protein structure model validation using DAQ Score

结构建模何时会出错?基于DAQ评分的PDB规模蛋白质结构模型验证分析

阅读:1

Abstract

Errors in structure modeling occur probably more frequently than one might think in cryo-EM structure determination, particularly when the map resolution is not very high. To this end, we have developed DAQ (Deep-learning-based Amino acid-wise model Quality) score, a machine learning-based method for detecting structural outliers in protein models. DAQ employs deep neural networks to analyze local density features of amino acids and atoms, assessing the likelihood of correct residue modeling (Terashi et al., Nat. Methods, 2022). Here, we present a large-scale analysis of over 10,000 protein structure models from the Protein Data Bank (PDB), revealing common trends in modeling errors. Our findings provide insights into systematic inaccuracies and guide improvements in structure validation. To facilitate broader accessibility, DAQ assessment results are available in the DAQ-DB database (https://daqdb.kiharalab.org) and integrated into PDBj (Nakamura et al., Nat. Methods, 2023). Furthermore, we introduce DAQ-Refine, an automated refinement protocol designed to correct errors detected by DAQ (Terashi et al., Acta Cryst. D, 2023).

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。