MARCEL

MARCEL是由加州大学洛杉矶分校等机构创建的分子构象集合学习基准数据集，包含四个子数据集，覆盖了包括有机催化剂和过渡金属催化剂在内的化学多样性分子和反应级别的属性。数据集通过密度泛函理论（DFT）计算，强调了构象集合的热力学平均属性。MARCEL旨在全面评估从构象集合中学习分子属性的潜力，并提出了两个明确的构象集合学习策略，以改善3D分子表示学习模型的性能。该数据集的应用领域包括分子属性预测和新型药物候选设计，旨在解决分子灵活性在模型中的表示问题。

MARCEL is a benchmark dataset for molecular conformational ensemble learning, created by the University of California, Los Angeles (UCLA) and other institutions. It includes four sub-datasets, covering chemically diverse molecules including organic and transition metal catalysts, as well as reaction-level properties. The dataset is developed based on density functional theory (DFT) calculations, and places emphasis on the thermodynamic average properties of conformational ensembles. MARCEL aims to comprehensively evaluate the potential of learning molecular properties from conformational ensembles, and proposes two explicit conformational ensemble learning strategies to enhance the performance of 3D molecular representation learning models. Its application fields cover molecular property prediction and novel drug candidate design, and it is intended to address the challenge of representing molecular flexibility in models.