BayesBind

BayesBind数据集是由北卡罗来纳大学教堂山分校的研究团队创建，用于评估虚拟筛选模型的性能。该数据集包含14个验证目标和11个测试目标，平均每个目标有297个已知活动，旨在解决机器学习模型在结构不相似的蛋白质目标上的评估问题。数据集通过从BigBind验证和测试集中选择目标，并确保随机化合物与活性分子来自同一化学空间，从而提高了评估的准确性和实用性。

The BayesBind dataset was created by a research team at the University of North Carolina at Chapel Hill to evaluate the performance of virtual screening models. It includes 14 validation targets and 11 test targets, with an average of 297 known active compounds per target. This dataset aims to address the challenge of evaluating machine learning models across structurally dissimilar protein targets. Specifically, targets were selected from the BigBind validation and test sets, with measures taken to ensure that random compounds and active molecules originate from the same chemical space, thereby enhancing the accuracy and practicality of the evaluation.