ABDataCorpus

# Antibody Extraction Dataset ## Dataset Overview 从43,580篇专利/文献中自动提取的抗体功能数据。 | Metric | Value | |--------|-------| | Total Documents | 43,583 | | Documents with Results | 37,587 | | Valid Document Ratio | 86.24% | | Selected Subset | 2,000 | | Total Antibodies | 15,990 | | High Quality Antibodies | 3,214 | ## File Structure ``` huggingface_ready/ ├── dataset_info.json # 数据集元信息 ├── folder_index.json # 文件夹索引信息 ├── high_quality_antibodies.json # 高质量抗体数据（有序列+亲和力+靶点） ├── antibodies_part_1.json # 抗体数据块1 ├── antibodies_part_2.json # 抗体数据块2 └── ... ``` ## Data Schema 每个抗体记录包含以下字段（可选）： - `Antibody_Name`: 抗体名称 - `Antibody_Type`: 抗体类型 (mAb, scFv, Fab, VHH, etc.) - `Antibody_Isotype`: 抗体亚型 - `source`: 来源物种 - `Target_Name`: 靶点名称 - `Target_Type`: 靶点类型 - `Epitope`: 表位信息 - `Mechanism_of_Action`: 作用机制 - `Experiment`: 实验方法 - `Binding_Kinetics_KD`: KD值 - `Binding_Kinetics_kon`: kon值 - `Binding_Kinetics_koff`: koff值 - `Binding_EC50`: EC50值 - `CDRH3_Sequence`: CDRH3氨基酸序列 - `vh_sequence_aa`: 重链可变区序列 - `vl_sequence_aa`: 轻链可变区序列 - `Structure`: 结构信息 - `Thermal_Stability_Tm`: 热稳定性Tm值 - `In_Vivo_Half_Life`: 体内半衰期 - `In_Vivo_Efficacy`: 体内有效性 - `Cross_Reactivity`: 交叉反应性 - `Quantitative_Metric`: 其他定量指标 - `Reference_Source`: 参考文献来源 ## High Quality Antibodies High Quality Antibodies满足： - 有序列信息 (VH/VL/CDRH3 任一) - 有亲和力数据 (KD/EC50/kon/koff 任一) - 有靶点信息 ## Usage ```python import json # 加载元数据 with open("dataset_info.json", "r") as f: info = json.load(f) # 加载所有抗体数据 all_antibodies = [] import glob for f in glob.glob("antibodies_part_*.json"): with open(f, "r") as f: all_antibodies.extend(json.load(f)) # 加载高质量抗体 with open("high_quality_antibodies.json", "r") as f: high_quality = json.load(f) ``` ## Source - Data Extracted: April/May 2026 - Extraction Method: ABCurator (multi-agent system) - Dataset License: [Add Your License] ## Citation [Add Citation Information]