Molecular substructure graph attention network for molecular property identification in drug discovery. (August 2022)

• Record Type: Journal Article
• Authors: Ye, Xian-bin; Guan, Quanlong; Luo, Weiqi; Fang, Liangda; Lai, Zhao-Rong; Wang, Jun
• Is Part Of: Pattern recognition. Volume 128(2022)
• Publication Date: 2022-08
• Subjects: Molecular substructure -- Graph attention -- Molecular property identification; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108659
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose to use a structural feature extraction scheme including 3 types of features (raw + tree decomposition + ECFP). We design a framework including several graph attention convolutional (GAC) blocks and deep neural network (DNN) blocks to process the above structural features. We design a readout block based on gated recurrent units (GRU). The readout blocks collaborate with the GAC blocks to obtain molecular embeddings. We visualize molecules and mark the important atoms with attention scores of MSSGAT, which can be a good reference for subsequent drug development. Abstract: Molecular machine learning based on graph neural network has a broad prospect in molecular property identification in drug discovery. Molecules contain many types of substructures that may affect their properties. However, conventional methods based on graph neural networks only consider the interaction information between nodes, which may lead to the oversmoothing problem in the multi-hop operations. These methods may not efficiently express the interacting information between molecular substructures. Hence, We develop a Molecular SubStructure Graph ATtention (MSSGAT) network to capture the interacting substructural information, which constructs a composite molecular representation with multi-substructural feature extraction and processes such features effectively with a nested convolution plus readout scheme. We evaluate the performance of our model on 13 benchmark data sets, in which 9 data sets are from the ChEMBL data base and 4 are the SIDER, BBBP, BACE, and HIV data sets. Extensive experimental results show that MSSGAT achieves the best results on most of the data sets compared with other state-of-the-art methods.