Much of computational materials science has focused on fast and accurate forward predictions of materials properties, for example, given a molecular structure predict its electronic properties. This is achieved with first principles calculations and more recently through machine learning approaches, since the former is computation-intensive and not practical for high-throughput screening. Searching for the right material for any given application, though follows an inverse path—the desired properties are given and the task is to find the right materials. Here we present a deep learning inverse prediction framework, Structure Learning for Attribute-driven Materials Design Using Novel Conditional Sampling (SLAMDUNCS), for efficient and accurate prediction of molecules exhibiting target properties. We apply this framework to the computational design of organic molecules for three applications, organic semiconductors for thin-film transistors, small organic acceptors for solar cells and electrolyte additives with high redox stability. Our method is general enough to be extended to inorganic compounds and represents an important step in deep learning based completely automated materials discovery.

许多计算材料科学都将注意力集中在对材料特性的快速，准确的正向预测上，例如，假设分子结构能够预测其电子特性。这是通过第一性原理计算实现的，最近又通过机器学习方法实现，因为前者的计算量很大，并且对于高通量筛选不切实际。为任何给定的应用程序寻找合适的材料，尽管遵循相反的路径-给出了所需的属性，任务是找到合适的材料。在这里，我们提出了一种深度学习逆向预测框架，即使用新型条件采样（SLAMDUNCS）进行属性驱动材料设计的结构学习，以有效且准确地预测具有目标特性的分子。我们将此框架应用于三种应用的有机分子的计算设计，用于薄膜晶体管的有机半导体，用于太阳能电池的小型有机受体以及具有高氧化还原稳定性的电解质添加剂。我们的方法足够通用，可以扩展到无机化合物，并且代表了基于深度学习的全自动材料发现中的重要一步。