appliedscientific/cardiosafe-benchmark

CardioSafe离子通道基准数据集是一个用于心脏离子通道活性多任务预测的基准数据集，包含经过整理的多任务离子通道标签和Tanimoto控制的拆分，以及用于《CardioSafe: 心脏离子通道活性多任务预测与反向泄漏审计基准测试》（Jovanović等人，2026年）的补充材料。该数据集旨在支持药物发现和化学信息学中的QSAR（定量结构-活性关系）研究，特别关注心脏毒性相关的离子通道，如hERG、Nav1.5、Cav1.2和IKs。数据集提供多个配置（如tan70_v1_0、tan60_v1_0等），每个配置包含训练、验证和测试拆分，数据以Parquet格式存储。标签列包括回归任务（如pIC50值）和分类任务（如阻断剂二元标签），并采用NaN稀疏表示。数据集基于ChEMBL 36和hERG Central初级筛选数据，经过药理学感知的整理策略，并包含Tanimoto相似性控制的拆分以确保数据独立性。The CardioSafe ion-channel benchmark is a curated multi-task ion-channel dataset with Tanimoto-controlled splits and supplementary materials for CardioSafe: multi-task prediction of cardiac ion channel activity with reverse-leak audited benchmarking (Jovanović et al., 2026). It is designed for QSAR (Quantitative Structure-Activity Relationship) studies in drug discovery and chemistry, focusing on cardiotoxicity-related ion channels such as hERG, Nav1.5, Cav1.2, and IKs. The dataset offers multiple configurations (e.g., tan70_v1_0, tan60_v1_0, etc.), each with train, validation, and test splits stored in Parquet format. Label columns include regression tasks (e.g., pIC50 values) and classification tasks (e.g., binary blocker labels), represented in a NaN-sparse manner. The dataset is derived from ChEMBL 36 and the hERG Central primary screen, curated with a pharmacology-aware policy, and features Tanimoto similarity-controlled splits to ensure data independence.