Research On Literature Based Discovery Methods In Biomedical Literature

Biomedical literature is currently the most important resource in biomedical research field.Nowadays,the amount of biomedical literatures is growing at an explosive speed,and much useful knowledge is yet undiscovered in the literature.Literature-based discovery(LBD)method focuses on discovering new knowledge from published biomedical literature,which has been successfully applied to the field of drug discovery,side effect prediction and drug mechanism of action,etc.In this dissertation,several LBD methods are proposed for discovering hidden knowledge from literature.There are mainly two types of LBD methods-ABC model based or AnC model based methods-for discovering hidden knowledge from a given set of biomedical literature.Based on the two types of the LBD methods,the main contents of this dissertation include:(1)A new ABC model based LBD method is proposed.The ABC model is currently the most widely used LBD framework,which infers the relationship between indirectly connected entities through a single co-occurring intermediate entity.Discovery pattern based method is currently the most effective ABC model based LBD method.The method discovers implicit knowledge through discovery patterns which are defined by domain experts.Due to the discovery patterns need to be manually defined,this method is not suitable for solving different types of LBD tasks.In this dissertation,an enhanced semantic discovery pattern based method is proposed for solving this problem.First,semi-supervised learning method is adopted to train two types of relation extraction models,then the two models construct the enhanced discovery pattern which is used to discover new knowledge from literature.Compared to traditional discovery pattern based method,the discovery pattern of our method is automatically obtained.In addition,the experimental results show our method achieves better performance.(2)An AnC model based LBD method is proposed.The AnC-based LBD methods can exploit complex implicit relationship between drugs and diseases from literature.The limitation of the existing methods is that these methods are mainly used to explain the relationship between drugs and diseases,there is no method could find potential drugs for a given disease.In this dissertation,two methods are presented to solve this problem:On the one hand,a semantic type path based method is proposed.The method first constructs a biomedical knowledge graph,and then uses the semantic type distribution of entities in the knowledge graph to discover implicit knowledge from literature.For a given disease of interest,the method can not only find its candidate therapeutic drugs but also give corresponding drug targets.On the other hand,a graph embedding based deep learning method is proposed for providing detailed explanations for drug-disease associations.The LBD task is considered as a sequential data analyzing problem in this method.The experimental results show that the method could provide detailed explanations for drug-disease associations.(3)An abstract selecting method for LBD is proposed.All existing LBD methods mainly focus on mining implicit knowledge from a given set of biomedical literature,regardless of obtaining the set of literature.In this dissertation,an abstract selecting method for LBD is proposed.This method selects abstracts by using the topic distribution of connected abstracts.The experimental results show that the proposed method can effectively select a small number of highly relevant abstracts for LBD,which improves the effectiveness of current LBD methods.