In recent years, after gaining enormous success in many areas, graph neural networks (GNN) have been applied in e-commerce, biomedicine, molecule pharmaceuticals, and astrophysics and received extensive attention from academia and industry as a key technology featuring intelligence, scale effects, high efficiency, low costs innovation.

Recently, Zhuque, the Lab's intelligent graph computing platform, claimed the champion of the Open Graph Benchmark (OGB) Challenge, a world-class competition for graph learning benchmarks. It outperformed the results of nearly one year ago, with an absolute advantage over well-known tech companies and universities, including Tencent, Shanghai Jiao Tong University, University of Notre Dame, and Cornell University. By winning the trophy, the Lab has displayed its strength in the GNN field.

Established by world-leading scientists, OGB is a collection of globally recognized benchmark datasets for graph learning that were made publicly available at the Conference on Neural Information Processing Systems (NeurIPS) in 2019. The datasets consist of multiple tasks, such as node property prediction, edge property link prediction, and graph property prediction. The competition is dubbed the “ImageNet” in graph learning for its high quality, large scale, complex scenarios, and challenging tasks. The Lab's platform won first place in the ogbl-ddi dataset, a subset of edge property link prediction. The task was to accurately predict interacting drug pairs from over 1.3 million drug-drug interactions. The champion title in the OGB Challenge is the proof of Zhuque’s leading position in machine learning.

Zhejiang Lab's Zhuque team won first place for drug interaction prediction

According to the team, AI-assisted drug development will remarkably increase the R&D efficiency and odds of success and reduce R&D and experiment expenses. It is a new hotspot in the fields of molecular property prediction, protein structure prediction, drug-target affinity prediction, etc. 

The team, led by CHEN Hongyang, Senior Research Expert and Deputy Director of ZJ Lab's Research Center for Graph Computing, has accumulated rich experience in machine learning and graph computing since long ago and established Zhuque, an intelligent graph computing platform. It is a large and efficient distributed memory-based graph computing platform that supports graph analysis algorithms and GNN learning based on Pytorch, Tensorflow and Mindspore. It supports both computing in memory and software-hardware co-optimization for operators. It also aggregates the computing capacities for multi-source heterogeneous data, such as graph analysis algorithms, graph interactive query, graph representation learning, and graph generation learning, so that computing for drug development, bio breeding and other scientific studies can be performed on this platform.

Zhuque Intelligent Graph Computing Platform

In recent years, the Lab's intelligent graph computing team has explored intelligent computing, data mining, deep learning, and other areas and achieved a series of research results. They have published over 50 papers in CCF-A journals and conferences, including IJCAI, SIGMOD, KDD, ICML, NIPS, AAAI and TKDE. And team members have won the following awards: the CCF Scientific and Technological Progress Award, CIKM AnalytiCup Champion in 2019, WWW 2021 Workshop Best Paper Award, and IJCAI 2021 Video Competition Award. In the future, the team will scale up its research in graph computing and AI for Science, achieve more breakthroughs in AI+X, and further play the role of intelligent graph computing in industrial transformation and scientific innovation.