Scholarlink

The paper describes a new approach to use Graph Neural Networks (GNNs) to predict protein-protein interactions (PPIs). This method tries to improve the accuracy and speed of PPI predictions by using the expressivity of graphs and language models. In particular, the paper describes two graph-based methods, GCN-based and GAT-based. Pre-trained language models were leveraged for feature extraction, which yielded better results than standard encoding methods like one-hot encoding. The results show that the GATv2 model performed better than both GCN and GAT with an accuracy of 98.31%, an f-score of 98.85%, a sensitivity of 99.01%, and a precision of 98.68%. A comparison with other works based on similar PPI datasets showed that the GATv2 approach is better based on several metrics.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Protein-Protein Interaction Prediction Using Graph Neural Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Othmane Boumya,
  • Hamza Hraiche,
  • Kaouter Karboub

摘要

The paper describes a new approach to use Graph Neural Networks (GNNs) to predict protein-protein interactions (PPIs). This method tries to improve the accuracy and speed of PPI predictions by using the expressivity of graphs and language models. In particular, the paper describes two graph-based methods, GCN-based and GAT-based. Pre-trained language models were leveraged for feature extraction, which yielded better results than standard encoding methods like one-hot encoding. The results show that the GATv2 model performed better than both GCN and GAT with an accuracy of 98.31%, an f-score of 98.85%, a sensitivity of 99.01%, and a precision of 98.68%. A comparison with other works based on similar PPI datasets showed that the GATv2 approach is better based on several metrics.