CESPED

CESPED是一个专为Cryo-EM中的监督姿态估计设计的新数据集，由牛津大学统计系和Astex Pharmaceuticals共同创建。数据集包含多种类型的分子，包括可溶性和膜结合分子，对称和非对称复合物，分辨率从5 Å到3.2 Å不等。CESPED旨在通过提供一个标准化的基准，促进深度学习方法在粒子处理中的应用，特别是在监督姿态估计模型上的改进。数据集的创建过程涉及从EMPIAR中手动搜索包含至少200,000个粒子的条目，并通过Relion版本4的自动精炼程序进行数据处理，确保重建体积的分辨率接近文献报告的值。CESPED的应用领域主要集中在Cryo-EM图像处理，特别是提高姿态估计的效率和准确性，从而推动Cryo-EM技术在结构生物学中的应用。

CESPED is a novel dataset tailored for supervised pose estimation in Cryo-EM, jointly created by the Department of Statistics, University of Oxford and Astex Pharmaceuticals. The dataset covers a wide range of molecular species, including soluble and membrane-bound molecules, symmetric and asymmetric complexes, with resolution values ranging from 5 Å to 3.2 Å. The core goal of CESPED is to promote the application of deep learning approaches in particle processing, especially the improvement of supervised pose estimation models, by offering a standardized benchmark. The dataset construction involved manually searching the EMPIAR database for entries containing at least 200,000 particles, followed by data processing using the automatic refinement pipeline of Relion version 4, to ensure that the resolution of the reconstructed volumes is consistent with values reported in existing literature. CESPED is primarily targeted at Cryo-EM image processing applications, with the aim of enhancing the efficiency and accuracy of pose estimation, thereby advancing the application of Cryo-EM technology in structural biology.