{"title":"An Innovative Multilevel Line Graph Attention Network for Predicting Molecular Properties","authors":"Yeling Zhang, Huancong Shi, Zhihui Chen, Dongfang Wu, Linhua Jiang","doi":"10.1109/INSAI56792.2022.00044","DOIUrl":null,"url":null,"abstract":"Predicting molecular properties is a fundamental task of quantum chemistry and a prerequisite for subsequent research, such as compound discovery and drug design. Recent studies show that, graph neural networks are more effective for this task. By analyzing the composition of molecular energy, we designed a multilevel line graph method for graph data generation, which was experimentally verified to be more capable of representing molecules. Furthermore, we designed an innovative multi-task graph neural network based on graph attention network to predict molecular properties by learning the features embedded in multilevel line graph. Experiments show that our method has higher accuracy in predicting energy properties. Specifically, our method improves performance when trained with fewer samples, which has a great significance for practical applications with sparse samples.","PeriodicalId":318264,"journal":{"name":"2022 2nd International Conference on Networking Systems of AI (INSAI)","volume":"283 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 2nd International Conference on Networking Systems of AI (INSAI)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/INSAI56792.2022.00044","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Predicting molecular properties is a fundamental task of quantum chemistry and a prerequisite for subsequent research, such as compound discovery and drug design. Recent studies show that, graph neural networks are more effective for this task. By analyzing the composition of molecular energy, we designed a multilevel line graph method for graph data generation, which was experimentally verified to be more capable of representing molecules. Furthermore, we designed an innovative multi-task graph neural network based on graph attention network to predict molecular properties by learning the features embedded in multilevel line graph. Experiments show that our method has higher accuracy in predicting energy properties. Specifically, our method improves performance when trained with fewer samples, which has a great significance for practical applications with sparse samples.