The current tasks of biomedical relation extraction mainly focus on the extraction of binary relations in single sentences, such as protein-protein interaction (PPI), chemical-protein interaction (CPI) and drug-drug interaction (DDI) [1–3]. It is crucial for biomedical relation extraction to automatically construct a knowledge graph, which supports a variety of downstream natural language processing (NLP) tasks such as drug discovery [4]. An obvious problem is that as the biomedical literature continues to grow, there is a large number of biomedical entities whose binary relations exist not only in a single sentence but also in cross-sentences. In addition, the relations between entities are not merely a binary relation but may also be an n-ary relation. Consider the following example: the relations between drugs, genes and mutations. “The deletion mutation on exon 19 of theEGFRgene was present in 16 patients, while theL858Epoint mutation on exon 21 was noted in 10. All patients were treated withgefitiniband showed a partial response.”. The message conveyed by the two sentences is that there is a reaction between the three bold entities. As the biomedical literature contains a wealth of drug-gene-mutation relations, how to quickly and accurately identify the drug-gene-mutation relations is particularly important in the treatment of precision medicine [5].
展开▼
