MMR: A Multi-view Merge Representation model for Chemical-Disease relation extraction

Comput Biol Chem. 2024 Apr 3;110:108063. doi: 10.1016/j.compbiolchem.2024.108063. Online ahead of print.ABSTRACTChemical-Disease relation (CDR) extraction aims to identify the semantic relations between chemical and disease entities in the unstructured biomedical document, which provides a basis for downstream tasks such as clinical medical diagnosis and drug discovery. Compared with general domain relation extraction, it needs a more effective representation of the whole document due to the specialized nature of texts in the biomedical domain, including the biomedical entity and entity-pair representation. In this paper, we propose a novel Multi-view Merge Representation (MMR) model to thoroughly capture entity and entity-pair representation of the document. First, we utilize prior knowledge and a pre-trained transformer encoder to capture entity semantic representation. Then we employ the U-Net layer and Graph Convolution Network layer to capture global entity-pair representation. Finally, we get a specific merged representation for each entity pair to be classified. We evaluate our model on the CDR dataset published by the BioCreative-V community and achieve a state-of-the-art result.PMID:38613989 | DOI:10.1016/j.compbiolchem.2024.108063
Source: Computational Biology and Chemistry - Category: Bioinformatics Authors: Source Type: research