B-XAIC

B-XAIC数据集由爱丁堡大学和雅盖隆大学数学与计算机科学学院的研究人员创建，旨在为图神经网络的可解释人工智能（XAI）提供基准。该数据集包含5万个小分子和7个不同的任务，每个任务都配有真实标签和相应的解释，使得基于准确性的度量方法可以直接应用并可靠地评估。数据集来源于ChEMBL 35数据库，通过加权采样方法避免了巨大的类不平衡问题，平均图大小为34.56。数据被随机分为训练、验证和测试集，使用8-1-1的比例。数据集包含每个化合物的二进制任务标签和两个解释标签集，一个用于检测到的模式中的原子，一个用于边缘。B-XAIC数据集提供了对XAI方法进行评估的宝贵资源，有助于深入了解XAI的可靠性，并促进更可靠和可解释模型的发展。

The B-XAIC dataset was created by researchers from the University of Edinburgh and the Faculty of Mathematics and Computer Science of Jagiellonian University, with the goal of providing a benchmark for explainable artificial intelligence (XAI) in graph neural networks. This dataset includes 50,000 small molecules and 7 distinct tasks, each paired with ground-truth labels and corresponding explanations, allowing accuracy-based evaluation metrics to be directly and reliably applied. Derived from the ChEMBL 35 database, the dataset uses weighted sampling to mitigate severe class imbalance issues, with an average graph size of 34.56. The data is randomly split into training, validation and test sets at an 8-1-1 ratio. The dataset contains binary task labels for each compound and two sets of explanation labels: one for atoms in detected patterns, and the other for edges. The B-XAIC dataset offers a valuable resource for evaluating XAI methods, enabling in-depth insights into the reliability of XAI and advancing the development of more reliable and interpretable models.