单体香料的留香等级(持久性)预测对于调香技术的发展有着重要意义。传统方法易受主观判断影响并且存在复杂的模型决策。因此，文中提出从单体香料的三维层面进行研究，基于三维图卷积神经网络(3DGCN)模型构建一个单体香料的留香分类框架。3DGCN模型在预测留香等级的任务上，相较于二维研究方法表现出了更好的性能优势，尤其是在图聚合阶段使用set2set池化时，3DGCN模型的分类结果准确率为82.06%，精确度为82.34%，召回率为81.98%，F1值为82.18%。这项实验基于三维图卷积神经网络的单体香料留香分类框架（MFRC-3DGCN）为留香等级的预测提供新的数据考量维度，同时也为单体香料留香等级属性评估提供可靠工具。

The prediction of fragrance retention grades (persistence) of monomer flavors is of great importance for the development of perfumery technology. Traditional methods are susceptible to subjective judgments and complex model decisions. Therefore, the paper proposes to investigate the three-dimensional dimension of monomer flavors and construct a monomer flavors retention classification framework based on a three-dimensional graph convolutional network (3DGCN) model. The 3DGCN model shows better performance in the task of predicting the fragrance retention grades compared to 2D research methods, especially when using set2set pooling in the graph aggregation stage, the classification results of the 3DGCN model with 82.06% accuracy, 82.34% precision, 81.98% recall, and 82.18% F1 score. This experiment is based on a three-dimensional graphical convolutional network framework for monomer flavors retention classification (MFRC-3DGCN), which provides a new dimension of data consideration for the prediction of fragrance retention grades and a reliable tool for the evaluation of fragrance retention grades properties of monomer flavors.

TQ657 ??????

国家自然科学基金项目（面上项目，重点项目，重大项目）