TEDBench/afdb

TEDBench-AFDB（预训练语料库）是一个用于蛋白质结构预训练的数据集，包含来自Foldseek聚类的AlphaFold数据库（pLDDT > 80）的代表性蛋白质，用于在TEDBench基准中预训练MiAE（Masked Invariant Autoencoders）。TEDBench是一个大规模、非冗余的蛋白质折叠分类基准，基于Encyclopedia of Domains (TED)和Foldseek聚类的AlphaFold结构构建。数据集总共有749,679个结构（训练集742,183，验证集7,496），每个Foldseek序列相似性聚类有一个代表性结构。数据包括名称、氨基酸序列、骨干坐标、pLDDT置信度分数、残基索引和ESM标记序列ID，没有标签列，仅用于无监督预训练。

TEDBench-AFDB (pretraining corpus) is a dataset for protein structure pretraining, containing representative proteins from Foldseek-clustered AlphaFold Database (pLDDT > 80), used to pretrain MiAE (Masked Invariant Autoencoders) in the TEDBench benchmark. TEDBench is a large-scale, non-redundant benchmark for protein fold classification constructed from the Encyclopedia of Domains (TED) and Foldseek-clustered AlphaFold structures. The dataset has a total of 749,679 structures (train: 742,183, val: 7,496), with one representative structure per Foldseek sequence-similarity cluster. It includes columns for name, amino-acid sequence, backbone coordinates, pLDDT confidence scores, residue index, and ESM-tokenised sequence IDs, with no label column, intended only for unsupervised pretraining.