introvoyz041/qmof_project

这是一个用于金属有机框架（MOF）性质预测的预处理数据集，旨在训练图神经网络。数据集整合了五个源数据库：QMOF、ODAC23、hMOF、CoREMOF和MOSAEC，并针对三种模型架构（CGCNN、MGT和PMT）进行了优化。数据以多种格式提供，包括CIF结构文件与id_prop.csv目标文件的捆绑包（适用于直接CGCNN/MGT输入）、预计算的LMDB图数据库（可直接用于训练）以及原始上游源文件。目标变量在预处理时未固定，而是在启动模型时从id_prop.csv文件中选择。数据集覆盖了约20,372到160,000个MOF结构，涉及CO₂和H₂O吸附对、假设MOF和实验MOF等类型，适用于材料科学和量子材料研究。Pre-processed datasets for training graph neural networks on Metal-Organic Framework (MOF) property prediction. Covers five source databases — QMOF, ODAC23, hMOF, CoREMOF, and MOSAEC — prepared for three model architectures: CGCNN, MGT, and PMT. Each database is available in up to three forms: CIF structures bundled with id_prop.csv target files (ready for direct CGCNN/MGT input), pre-computed LMDB graph databases (ready for training), and raw upstream source archives. The target variable for training is selected at model launch from id_prop.csv files. The datasets include structures ranging from approximately 20,372 to 160,000 MOFs, covering CO₂ and H₂O adsorbate pairs, hypothetical MOFs, and experimental MOFs, for materials and quantum materials research.