Research On Drug-disease Interaction Prediction Based On Biomedical Semantic Data

Drug development is time-consuming and labor-intensive.Drug repositioning is an important way to solve these problems and drug-disease interaction prediction is a research hot topic in drug repositioning.The traditional methods were mainly based on text mining related methods of relation link prediction,but the potential mechanism of disease could not be considered.To process and analyze complex interactions in disease mechanism well,network models were applied for drug-disease interaction prediction studies.Although the existing interaction prediction methods based on network models consider the integrity of complex relations,they ignore the node differences in the network.At the same time,most models are drug centered.In addition,the evaluation part of most methods lacks validation and result evaluation of multiple diseases.To solve these problems,we firstly take the disease-centered pathogenic factor as a trail,comprehensively considers all genes and protein entities having a pathogenic semantic relation with disease.And propose drug-disease relation prediction method(QSRCPM)based on calculating the quantitative semantic relation.Considering the differences among network nodes,our paper proposes a causal contribution network model and a drug-disease interaction prediction method(CCNPM)based on this network model.This method proposes the concept of node contribution based on Page Rank to reflect the difference of nodes.On this basis,we propose a network node centrality algorithm(NC-Page Rank)based on node confidence and a drug-disease interaction prediction method(CCNNCPM)based on causal contribution network with node confidence.Meantime,our paper also designed sorting strategies for the above three methods,which are gene prioritization rank strategy and protein prioritization rank strategy.Finally,our paper utilizes biomedical literature dataset to carry on drug-disease interaction prediction experiments and evaluate experimental results on Parkinson's disease,Breast cancer and Alzheimer's disease.In experiment part,we respectively using the proposed three drug-disease interaction prediction methods: QSRCPM,CCNPM,and CCNNCPM.The experimental results show that the three methods show excellent performance in the drug-disease interaction prediction tasks,and the drug-disease interaction prediction method with disease-centered pathogenic factor as a trail has good versatility and expansibility.