关键词: 键解离能, 随机森林, iBonD, 原子类型, 化学键类型

Abstract:

1208 organic molecules containing C， H， O， N， and S were manually collected from the iBonD organic bond energy database， and the corresponding experimental bond dissociation energy values were recorded. Chemical bond type descriptors， heteroatomic count descriptors， and branch descriptors were proposed and combined with previously suggested atomic type descriptors to provide a more comprehensive description of the surrounding environment of the target chemical bond. The prediction models for bond dissociation energy were constructed using random forest， and the results show that the combination of the descriptors of atomic types and chemical bond types around the target chemical bond achieves the best prediction results， and satisfactory results were obtained without quantum chemistry assistance. Compared with the results in published literature， the predicted results herein are better than the corresponding results in the literature. In addition， an algorithm on the application domain was designed to assess the quality of prediction results in advance， and the training set and the test set were randomly re-partitioned to verify the stability of the model， as well as the feasibility of the model was evaluated by comparing it with a zero model.

Key words: Bond dissociation energy, Random forest, iBonD, Atom type, Bond type