First-principles-based exploration of chemical space deepens our understanding of chemistry and might help with the design of new molecules, materials or experiments. Due to the computational cost of quantum chemistry methods and the immense number of theoretically possible stable compounds, comprehensive in silico screening remains prohibitive. To overcome this challenge, we combine atom-in-molecule-based fragments, dubbed ‘amons’ (A), with active learning in transferable quantum machine learning (ML) models. The efficiency, accuracy, scalability and transferability of the resulting AML models is demonstrated for important molecular quantum properties such as energies, forces, atomic charges, NMR shifts and polarizabilities and for systems including organic molecules, 2D materials, water clusters, Watson–Crick DNA base pairs and even ubiquitin. Conceptually, the AML approach extends Mendeleev’s table to account effectively for chemical environments, which allows the systematic reconstruction of many chemistries from local building blocks.

基于第一原理的化学空间探索加深了我们对化学的理解，并可能有助于设计新的分子，材料或实验。由于量子化学方法的计算成本和理论上可能存在的大量稳定化合物的原因，全面的计算机筛查仍然令人望而却步。为了克服这一挑战，我们将基于分子中原子的片段（称为“原子”（A））与可转移量子机器学习（ML）模型中的主动学习相结合。对于重要的分子量子特性，例如能量，力，原子电荷，NMR位移和极化率，以及包括有机分子，2D材料，水团簇，Watson-Crick DNA在内的系统，证明了所得AML模型的效率，准确性，可扩展性和可转移性。碱基对，甚至泛素。