Pre-training graph neural networks for link prediction in biomedical networks

• Main Authors: LONG, Yahui, WU, Min, LIU, Yong, FANG, Yuan, KWOH, Chee Kong, LUO, Jiawei, LI, Xiaoli
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2022
• Subjects: Data Storage Systems; OS and Networks
• Online Access: https://ink.library.smu.edu.sg/sis_research/7158; https://ink.library.smu.edu.sg/context/sis_research/article/8161/viewcontent/BIBM.pdf
• Institution: Singapore Management University

Motivation: Graphs or networks are widely utilized to model the interactions between different entities (e.g., proteins, drugs, etc) for biomedical applications. Predicting potential links in biomedical networks is important for understanding the pathological mechanisms of various complex human diseases, as well as screening compound targets for drug discovery. Graph neural networks (GNNs) have been designed for link prediction in various biomedical networks, which rely on the node features extracted from different data sources, e.g., sequence, structure and network data. However, it is challenging to effectively integrate these data sources and automatically extract features for different link prediction tasks. Results: In this paper, we propose a novel pre-training model to integrate different data sources for link prediction in biomedical networks. First, we design expressive deep learning methods (e.g., CNN and GCN) to learn features for individual nodes from sequence and structure data. Second, we further propose a GCN-based encoder to effectively refine the features of nodes by modelling the dependencies among nodes in the network data. Third, the model is pre-trained based on graph reconstruction tasks. Extensive experiments have been conducted on two critical link prediction tasks, i.e., synthetic lethality (SL) prediction and drug-target interaction (DTI) prediction. Experimental results demonstrate that the features generated by our pre-training model can help to improve the performance and reduce the training time for existing GNN models. In addition, fine-tuning the pre-trained model to a specific task can also achieve the performance comparable to the state-of-the-art methods.