Interatomic Potentials for Small Molecule Radical Reactions

第一年：为CDT所有一年级学生成员提供一般培训活动。第2-4年：机器学习原子间势是原子建模中的新兴工具。它们弥补了精确但缓慢的量子力学（QM）方法与计算便宜但不令人满意的精确经典力场之间的差距。虽然它们已经被应用于模拟一系列材料和有机分子，但它们在模拟反应中的应用还有待探索。机器学习电位建立在由三维结构和能量和力组成的参考数据库上，用高精度的QM方法估计。因此，它们只适用于与力场所适用的结构不太相似的结构。因此，训练数据集的组成成为人们关注的焦点。最近的一项发展是利用不完美电位来探索化学空间，并利用收集到的数据来改进电位。在我们的案例中，挑战是确定如何以类似的方式收集小有机分子和自由基的数据，以解决我们试图解决的相对受限的问题。我们将缩小“小分子自由基模拟反应”的范围，重点关注与Cytochromes P450代谢相关的甲氧基自由基和类药物分子的吸氢反应。为了模拟反应，人们需要收集反应物和生成物平衡几何的代表性数据，以及从它们之间的反应路径中采样的数据。

Year 1: Generic training activities for all first-year student members of the CDT.Year 2-4: Machine-learned interatomic potentials are emerging tools in atomistic modelling. They bridge the gap between the accurate yet slow quantum-mechanical (QM) methods and computationally cheap, but not satisfyingly accurate classical force fields. Although they have been applied to modelling a range of materials and organic molecules, their use in modelling reactions is yet to be explored.ML potentials are built on reference databases consisting of three-dimensional structures and energies and forces estimated with a highly accurate QM methods. As a result they are only applicable to structures not too dissimilar to ones the force fields were fitted to. Consequently, a lot of attention is devoted to composing the training data sets. A recent development is to use an imperfect potential to explore the chemical space and improve the potential with collected data. The challenge in our case is to determine how to similarly gather data for small organic molecules and radicals for the relatively constrained problem we are trying to solve.We will narrow the scope of "modelling reactions of small molecules and radicals" by focusing on hydrogen abstraction reactions of methoxy radical and drug-like molecules, relevant to Cytochromes P450 metabolism. To model reactions, one would need to gather representative data for the equilibrium geometries of reactants and products as well as data sampled from the reaction path between them.