Simulated MeSH Hierarchical Dataset with Poincaré Hyperbolic Distances

This dataset simulates one million hierarchical relationships between biomedical concepts inspired by the MeSH vocabulary, capturing valid parent-child and unrelated pairs. Each entry contains unique IDs for parent and child concepts, their hierarchical tree numbers, a hyperbolic distance score representing their proximity in a Poincaré embedding space, and a binary label indicating whether the child is a true hierarchical descendant. Valid child nodes have tree numbers extending their parent’s and low hyperbolic distances (0.01–0.4), while invalid pairs have unrelated tree numbers and larger distances (0.8–2.0). This dataset supports research on hierarchical representation learning and classification in biomedical ontologies. Finally, the dataset is part of the larger project regarding oncological solving via computational biology in AURORA.本数据集受医学主题词表（MeSH）启发，模拟了一百万条生物医学概念间的层级关系，涵盖合法的父子概念对与无关概念对。每条数据均包含父子概念的唯一标识符、各自的层级树编号、用于表征二者在庞加莱嵌入（Poincaré embedding）空间中邻近程度的双曲距离得分，以及用于标识子概念是否为父概念真实层级后代的二元标签。合法的子概念节点的层级树编号为其父节点编号的延伸，且双曲距离较低（0.01–0.4）；而非法概念对的层级树编号互不相关，且双曲距离较高（0.8–2.0）。本数据集可用于支撑生物医学本体中的层级表征学习与分类任务相关研究。此外，本数据集隶属于AURORA项目的子课题，该课题围绕通过计算生物学手段解决肿瘤学相关问题展开。