DAEM: Deep attributed embedding based multi-task learning for predicting adverse drug–drug interaction. (1st April 2023)

• Record Type: Journal Article
• Authors: Zhu, Jiajing; Liu, Yongguo; Zhang, Yun; Chen, Zhi; She, Kun; Tong, Rongsheng
• Is Part Of: Expert systems with applications. Volume 215(2023)
• Publication Date: 2023-04-01
• Subjects: Adverse drug–drug interaction -- Attributed embedding -- Multi-task learning -- Non-linear property; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.119312
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Adverse drug–drug interaction (ADDI) is an important concern in pharmaceutical industry and becomes a leading cause of morbidity and mortality in public health. With the increasing accumulation of biochemical characteristics of drugs, many computational methods are proposed by exploiting multiple attributes of drugs for ADDI prediction. However, due to the high-dimensional and highly sparse spaces of the hand-designed attributes of drugs, it still remains a challenging issue for investigating a robust projection between attributes of drugs and their adverse interactions, which can benefit to revealing the non-linear properties of their adverse relationship for accurate ADDI prediction. In this paper, we propose a Deep Attributed Embedding based Multi-task (DAEM) learning model for ADDI prediction. In particular, two drug attributes, molecular structure and side effect, are adopted to model the adverse interactions among drugs and a deep neural network is designed to embed the hand-designed attributes into their low-dimensional spaces while preserving adverse relationship and modeling attribute dependence for learning the informative attribute representations and capturing the non-linear properties of drugs. Along this line, multi-task learning is performed for ADDI prediction by regarding the prediction of each ADDI as a regression task jointly with proper regularizations. Experimental results on real-world dataset demonstrate the effectiveness of DAEM when compared with thirteen baselines and its variants. Highlights: A model DAEM with adoption of two drug attributes is devised for ADDI prediction. Deep attributed embedding is designed to learn abstract representations of drugs. A tactics is devised to retain adverse relationship and model attribute dependence. Experimental results demonstrate the excellent performance achieved by DAEM.