AMGC is a multiple-task graph neutral network for epigenetic target profiling

Shukai Gu, Lingjie Bao, Yuwei Yang, Yihao Zhao, Henry Hoi Yee Tong, Liwei Liu, Huanxiang Liu, Tingjun Hou, Yu Kang

Research output: Contribution to journalArticlepeer-review


Accurate prediction of ligand-protein interactions is crucial in the initial stages of drug development, especially when focusing on intricate epigenetic targets. This study proposes AMGC, an adaptive multi-task graph conventional network with contrastive learning to predict the inhibitory activities of small molecules against a panel of 67 epigenetic targets. The results show that AMGC outperforms classic molecular fingerprint-based machine learning models and other graph neutral network (GNN) variant-based models in terms of predictive and generalization abilities. Additionally, owing to the advantages of multi-task learning on small datasets, the AMGC approach is well suited for epigenetic target fishing and surpasses the state-of-the-art consensus model ETP. The interpretability of AMGC is verified by highlighting the key atoms in the ligand that interact with surrounding residues. To facilitate easy access, we develop a user-friendly online server for the epigenetic target fishing service.

Original languageEnglish
Article number101850
JournalCell Reports Physical Science
Issue number3
Publication statusPublished - 20 Mar 2024


  • deep learning
  • epigenetic target profiling
  • explainability
  • multiple task


Dive into the research topics of 'AMGC is a multiple-task graph neutral network for epigenetic target profiling'. Together they form a unique fingerprint.

Cite this