chimera-bench
收藏资源简介:
CHIMERA-Bench v1.0 是一个用于表位特异性抗体CDR序列-结构协同设计的统一基准数据集。该数据集包含2,922个复合物,其中2,721个具有PDB结构,2,941个预计算的特征文件(.pt格式)。数据集提供了三种分割方式(表位组、抗原折叠、时间序列),并支持IMGT和Chothia编号方案。每个复合物特征文件包含序列、坐标、注释、编号和表面特征等详细信息。数据集适用于抗体设计、蛋白质结构预测和生物分子设计等任务,并提供了11种基准方法和6种范式的评估结果。数据以PyTorch张量和PDB文件形式存储,包含元数据、分割信息和结构文件。
CHIMERA-Bench v1.0 is a unified benchmark dataset for sequence-structure co-design of complementarity-determining region (CDR) sequences of epitope-specific antibodies. This dataset includes 2,922 complexes, among which 2,721 have resolved PDB structures, and 2,941 pre-computed feature files in .pt format. Three splitting strategies (epitope group, antigen fold, and temporal split) are provided, and the dataset supports both IMGT and Chothia numbering schemes. Each complex feature file contains detailed information such as sequences, coordinates, annotations, numbering details, and surface features. This dataset is applicable to tasks including antibody design, protein structure prediction, and biomolecular design, and provides evaluation results for 11 benchmark methods and 6 design paradigms. The data is stored in the form of PyTorch tensors and PDB files, covering metadata, splitting information, and structural files.




