标题： 理解与提升机器学习活化能预测器的迁移能力 显示英文标题

标题： Understanding and improving transferability in machine-learned activation energy predictors

摘要： 反应性质的计算是在化学反应发现过程中的一项挑战任务。机器学习（ML）方法在加速活化能和反应焓的电子结构预测中发挥着重要作用，并且是实现大规模自动化反应网络发现的关键因素，其中涉及$>10^3$反应。不幸的是，现有机器学习模型的预测准确性尚未达到所需水平，无法在整个可能的化学反应空间中启用后续的动力学模拟，这些模拟甚至在定性上也与实验动力学相一致。在这里，我们全面评估了机器学习的反应模型中预测失败的根本原因。基于反应物和产物结构之间差异指纹的模型尽管提供了良好的分布内预测，但缺乏迁移性。这导致了关于化学反应背景和机制的信息显著丢失。我们提出了一种卷积机器学习模型，该模型使用以原子为中心的量子化学描述符和近似的过渡态信息。包括后者可以提高非分布内基准反应的迁移性，更充分地利用训练数据所跨越的有限化学反应空间。该模型还提供了活化能和反应焓的原子级别贡献，这些贡献为合理化反应性提供了一个有用的解释工具。 显示英文摘要

摘要： The calculation of reactive properties is a challenging task in chemical reaction discovery. Machine learning (ML) methods play an important role in accelerating electronic structure predictions of activation energies and reaction enthalpies, and are a crucial ingredient to enable large-scale automated reaction network discovery with $>10^3$ reactions. Unfortunately, the predictive accuracy of existing ML models does not yet reach the required accuracy across the space of possible chemical reactions to enable subsequent kinetic simulations that even qualitatively agree with experimental kinetics. Here, we comprehensively assess the underlying reasons for prediction failures within a selection of machine-learned models of reactivity. Models based on difference fingerprints between reactant and product structures lack transferability despite providing good in-distribution predictions. This results in a significant loss of information about the context and mechanism of chemical reactions. We propose a convolutional ML model that uses atom-centered quantum-chemical descriptors and approximate transition state information. Inclusion of the latter improves transferability for out-of-distribution benchmark reactions, making greater use of the limited chemical reaction space spanned by the training data. The model further delivers atom-level contributions to activation energies and reaction enthalpies that provide a useful interpretational tool for rationalizing reactivity.