THEMol (Torsion, Hessian, Energy of Molecules)
收藏资源简介:
THEMol数据集是由字节跳动团队构建的大规模开源量子力学性质集合,专注于闭壳层有机分子,涵盖多达50个重原子。该数据集内容极为丰富,包含超过300万优化几何结构的Hessian矩阵子集、近1亿约束几何结构的扭转扫描子集,以及总计约30亿次DFT计算的弛豫轨迹子集,数据来源整合了UniChem等公共数据库及专有集合。其创建过程通过系统的分子片段化、质子化状态枚举、构象生成及B3LYP-D3(BJ)/DZVP理论级别的量子化学计算完成,并实施了严格的自旋态过滤、去重和几何收敛性质量控制。该数据集旨在为分子力学和机器学习势函数的开发提供坚实基础,特别适用于药物发现、电解质和离子液体等领域的分子势能面精确建模。
The THEMol dataset is a large-scale open-source collection of quantum mechanical properties developed by the ByteDance team, focusing on closed-shell organic molecules containing up to 50 heavy atoms. This dataset is highly comprehensive, including three subsets: the Hessian matrix subset with over 3 million optimized geometric structures, the torsion scan subset with nearly 100 million constrained geometric structures, and the relaxation trajectory subset involving approximately 3 billion total DFT calculations. Its data sources integrate public databases such as UniChem and proprietary collections. Its construction is completed through systematic molecular fragmentation, protonation state enumeration, conformation generation, and quantum chemical calculations at the B3LYP-D3(BJ)/DZVP theoretical level, with strict spin state filtering, deduplication, and geometric convergence quality control implemented. This dataset aims to provide a solid foundation for the development of molecular mechanics and machine learning potential functions, and is particularly suitable for accurate modeling of molecular potential energy surfaces in fields such as drug discovery, electrolytes, and ionic liquids.
数据集名称
THEMol dataset (Torsion, Hessian, Energy of Molecules)
数据集描述
THEMol是一个面向闭壳层有机分子(最多含50个重原子)的大规模开源量子力学性质数据集。
主要属性
- 覆盖元素:涵盖12种必需元素。
- 分子架构:覆盖与药物发现、电解质、离子液体等领域相关的多样化分子结构。
数据集子集
数据集由五个不同的子集组成:
- Hessian 子集:包含超过300万个带有Hessian矩阵的松弛几何结构。
- TorsionScan 子集:包含近1亿个带有能量和力的约束松弛几何结构,提供详尽的构象采样(包括环内和环外扭转发散扫描)。
- HessianRelax 子集:松弛轨迹子集。
- TorsionScanRelax 子集:松弛轨迹子集。
- MBIS 子集:包含通过最小基组迭代Stockholder划分方案计算的电子密度衍生的原子多极矩。
数据规模
- 总计包含约30亿次DFT计算。
- 包含优化的几何结构、松弛轨迹以及衍生的分子性质。
应用价值
该数据集旨在赋能高精度、高迁移性分子势能模型的开发。




