Protein-Protein Interaction Extraction based on Multi-feature Fusion and Entity Enhancement

Relationships between protein entities are a crucial component of the biomedical knowledge system. However, this information is often embedded in unstructured text, making its automatic extraction and conversion into structured data essential for biomedical knowledge mining and downstream applications, such as drug discovery, disease research, and precision medicine. Accurate extraction of protein-protein interactions not only enhances the understanding of biological processes but also facilitates the development of computational models for predicting protein functions, identifying potential therapeutic targets, and understanding disease mechanisms at the molecular level. In this study, we propose a novel approach aimed at extracting relational information between protein entities. The deep semantics are effectively captured by combining contextual semantic information with detailed word-level features. Additionally, we introduce an enhanced protein entity recognition strategy, further improving the accuracy of protein entity relationship extraction. We evaluate the method on five standard datasets for protein-protein interaction (PPI) extraction. Experimental results demonstrate that the proposed model significantly outperforms existing approaches and achieves excellent performance in the PPI extraction task.