Research And Application On Relation Extraction Between Microbes And Related Entities Based On Literature Mining

With recent advances in biotechnology,people got a gradually deeper understanding of the microorganisms.Relevant studies have found that microbes in vivo play an important role in human health.From the application point of view,along with the pervasive use of antibiotics,their drug resistance is also starting to be gradually manifested.Many pathogenic microbes in the human body start to appear less sensitive or even insensitive to antibiotics.The time and human cost required to develop new antibiotics are very high,so how to utilize existing drugs for the treatment of drug-resistant microorganisms is also one of the focuses in the microbial field nowadays.With the development of biomedical technology,a large number of research results are recorded in the form of literature,which contains a wealth of biomedical knowledge.How to efficiently and quickly extract structured knowledge from unstructured literature is one of the current research hotspots.Many of biomedical knowledge can be represented by the correlations between biomedical entities,and entity relationship recognition for biomedical literature is the core task in the field of biomedical literature mining.The interaction between biomedical entities obtained by literature mining can provide important theoretical support for disease mechanism analysis and prediction,drug use and redirection.In this regard,in this paper,taking microbe as the main research object,from the perspective of literature mining,we study the use of deep learning technology to realize the automatic identification of microbe-disease interactions and microbe-drug interactions in biomedical literature.The main research content of this paper mainly includes the following aspects:(1)The literature mining process for microbes,diseases and drugs is designed.Firstly,aiming at the named entity recognition problem of microbes,integrated multiple microberelated databases and build a human-related microbe dictionary; Secondly,extracted the entity information of microbes,diseases,and drugs related to the human body from the biomedical literature,and extract the <microbe,sentence,disease/drug> triplet in the literature.(2)A method for extracting microbe-disease interactions based on transfer learning is proposed,and a microbe-disease relationship database platform for biomedical researchers is constructed.In order to extract the microbe-disease relationship with a deep learning model in the case of less annotated data,this paper defines the classifications of microbe-disease relationships,and constructs a high-quality gold-standard dataset annotated by humans and a classified silver standard dataset with partial noise,which are used as the target domain and the source domain for transfer learning,respectively.The relevant comparative experiments demonstrate the effectiveness of the transfer learning method on the task of extracting microbe-disease interactions.In addition,a visual microbe-disease interactions database platform MDIDB was constructed based on the predicted data,and part of the data was randomly selected for exemplification analysis.(3)A Bio BERT-based microbe-drug interactions extraction model MDr I-BERT is proposed,and the literature mining process is applied to the extraction of the microbedrug relationship.The relationship between microbes and drugs was defined from the perspectives of the influence of microbes on drugs and the effects of drugs on microbes,and a manually annotated dataset was constructed for the training,validation and testing of deep learning models.MDr I-BERT has good results on the microbe-drug interactions dataset.Finally,the prediction data is verified and analyzed,taking the treatment drugs of2019-n Co V as an example,to verify the effectiveness of the extraction of microbe-drug relationship based on literature mining.