{"title":"药物相互作用预测的负抽样变分图网络","authors":"Jiawei Xu","doi":"10.1109/AINIT59027.2023.10212734","DOIUrl":null,"url":null,"abstract":"Drug-drug interactions (DDIs) are crucial for pharmaceutical research. Drug combinations could have negative medication consequences that endanger patient safety and public health. Deep learning techniques, especially graph neural networks, have demonstrated their effectiveness in graph learning, and thus have been widely applied in predicting DDIs. However, existing methods are limited in two aspects: 1) They heavily rely on the validity of graph structure, therefore fail to perform robust learning from noisy data 2) They make assumptions that all unobserved graph edges are irrelevant. However, it is more rational to view them as unlabeled links instead of negative ones. To address these challenges, we formulated the DDIs prediction problem as a graph link prediction task and proposed to train graph neural networks with variational learning and structure-aware negative sampling. Extensive experiments showed that our approach achieved improved performance than multiple baselines. Importantly, we performed a case study to evaluate the quality of our model's novel predictions by performing a literature-based evaluation of new hits. Evaluation results offer concrete examples that reaffirmed the medical usefulness of our approach.","PeriodicalId":276778,"journal":{"name":"2023 4th International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT)","volume":"323 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2023-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Variational Graph Network with Negative Sampling for Drug Interaction Prediction\",\"authors\":\"Jiawei Xu\",\"doi\":\"10.1109/AINIT59027.2023.10212734\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Drug-drug interactions (DDIs) are crucial for pharmaceutical research. Drug combinations could have negative medication consequences that endanger patient safety and public health. Deep learning techniques, especially graph neural networks, have demonstrated their effectiveness in graph learning, and thus have been widely applied in predicting DDIs. However, existing methods are limited in two aspects: 1) They heavily rely on the validity of graph structure, therefore fail to perform robust learning from noisy data 2) They make assumptions that all unobserved graph edges are irrelevant. However, it is more rational to view them as unlabeled links instead of negative ones. To address these challenges, we formulated the DDIs prediction problem as a graph link prediction task and proposed to train graph neural networks with variational learning and structure-aware negative sampling. Extensive experiments showed that our approach achieved improved performance than multiple baselines. Importantly, we performed a case study to evaluate the quality of our model's novel predictions by performing a literature-based evaluation of new hits. Evaluation results offer concrete examples that reaffirmed the medical usefulness of our approach.\",\"PeriodicalId\":276778,\"journal\":{\"name\":\"2023 4th International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT)\",\"volume\":\"323 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2023-06-16\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2023 4th International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/AINIT59027.2023.10212734\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2023 4th International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/AINIT59027.2023.10212734","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Variational Graph Network with Negative Sampling for Drug Interaction Prediction
Drug-drug interactions (DDIs) are crucial for pharmaceutical research. Drug combinations could have negative medication consequences that endanger patient safety and public health. Deep learning techniques, especially graph neural networks, have demonstrated their effectiveness in graph learning, and thus have been widely applied in predicting DDIs. However, existing methods are limited in two aspects: 1) They heavily rely on the validity of graph structure, therefore fail to perform robust learning from noisy data 2) They make assumptions that all unobserved graph edges are irrelevant. However, it is more rational to view them as unlabeled links instead of negative ones. To address these challenges, we formulated the DDIs prediction problem as a graph link prediction task and proposed to train graph neural networks with variational learning and structure-aware negative sampling. Extensive experiments showed that our approach achieved improved performance than multiple baselines. Importantly, we performed a case study to evaluate the quality of our model's novel predictions by performing a literature-based evaluation of new hits. Evaluation results offer concrete examples that reaffirmed the medical usefulness of our approach.