A Molecular Toxicity Prediction Method Based on Knowledge Enhancement of Atomic Properties
Current deep learning-based methods for toxicity prediction of chemical molecules mainly utilize the string representation of molecules,but the existing string representation models ignore the knowledge of the properties of different atoms in molecules,which leads to the failure of learning models fully utilizing the domain knowledge.To address these problems,a method that explicitly introduces hydrogen atoms and enhances the knowledge of atomic properties using the Morgan fingerprint radius is proposed to enable the toxicity prediction model to learn the knowledge of the properties of atoms in chemical molecules.In the improved toxicity prediction model,a sequence of molecular Morgan fingerprint identifiers enhanced with hydrogen atoms and atomic property knowledge is used as input,and the radius feature of molecular Morgan fingerprint is additionally introduced in the embedding layer.To validate the effectiveness of the proposed method,the pre-trained model was fine-tuned and tested on the mainstream toxicity prediction dataset Tox21.The experimental results showed that the improved method achieved the best AUC scores on multiple channels compared with the existing molecular sequence-based chemical molecule toxicity prediction methods.
万方数据
维普
