Scholarlink

<p>Protein function prediction is a key step in understanding biological systems revealing disease mechanisms, designing drugs. In this study, a novel model SSIPIN is proposed for protein function prediction. The model uses three types of data including protein sequences, protein-protein interaction networks (PPIN) and protein tertiary structures. The embedded protein features are used as the node features of the PPIN, and the PPIN is processed using Graph Attention Networks (GAT) thus learning the information of proteins that have interactions. We also design a method called Geometric Vector Perceptron (GVP)-GraphTransformer to process the protein structure, which is capable of mining deeper protein structure information. The feature representations obtained from the above two methods are then fused for better protein function prediction. Experimental results show that the proposed method outperforms existing methods on multiple datasets, providing an effective prediction tool for protein research.</p>

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

Ssipin: sequence-structure integrated protein interaction network for protein function prediction

  • 期刊论文

  • 发表日期：

    2026年8月4日

查看全文
  • Shuai Li,
  • Mingxuan Li,
  • Mandong Hu,
  • Zhen Li

摘要

Protein function prediction is a key step in understanding biological systems revealing disease mechanisms, designing drugs. In this study, a novel model SSIPIN is proposed for protein function prediction. The model uses three types of data including protein sequences, protein-protein interaction networks (PPIN) and protein tertiary structures. The embedded protein features are used as the node features of the PPIN, and the PPIN is processed using Graph Attention Networks (GAT) thus learning the information of proteins that have interactions. We also design a method called Geometric Vector Perceptron (GVP)-GraphTransformer to process the protein structure, which is capable of mining deeper protein structure information. The feature representations obtained from the above two methods are then fused for better protein function prediction. Experimental results show that the proposed method outperforms existing methods on multiple datasets, providing an effective prediction tool for protein research.