Multi-view Similarity Network Fusion For Biomedical Entities Association Prediction

Biological entity association prediction is an important research direction in modern bioinformatics,exploring the association between biological entities like RNA,drug,medication and diseases is useful for the study of assorted complicated disease pathologies,identification of malady biomarkers,target-specific drug analysis,and disease prevention and treatment.Biological experiments to verify the association of biological entities are highly accurate but time-consuming and costly.In recent years,computational methods have been inmerged in a large num as efficient complementary means to identify associations of biological entities,therefore it's significance to present computational methods that can explore biological entity associations accurately and effectively.In this paper,we present a multi-view similarity network fusion method(Mv SFN)to predict unknown biomedical entity associations.The Mv SFN method first combines known biological entity association data,existing bioinformatic data and heterogeneous association data to construct biological entity similarity networks corresponding to each dataset.The method aims to expand the function domain of the fusion method by introducing a nonlinear fusion method,so as to select a better similar data fusion method for the fusion of the basic similar networks.Finally,this paper combines the ideas of classical inductive matrix complementation method and proposes a neural inductive matrix complementation method to accomplish entity feature projection and entity association prediction.Combining the nonlinear interaction information between entities,we improve the linear projection method in the traditional inductive matrix complementation method to nonlinear encoding,so as to obtain better prediction results.In this paper,multiple validation tools are used on multiple datasets to prove the superiority of the method.The experimental datasets include two mi RNA-disease association datasets and five Lnc RNA-disease association datasets.The AUC results obtained by the global 5FCV method on the first five major data sets were0.9383+/-0.00062,0.9328+/-0.00031,0.9243+/-0.00018,0.9192+/-0.00085,0.9452+/-0.00062;the global LOOCV method The AUC results obtained were 0.959,0.954,0.931,0.930,0.940;the AUC results obtained by local 5FCV were 0.870,0.869,0.881,0.884,0.885.This paper also conducts case study for kinds of complex human diseases with the aim of further validating the effectiveness of the method.This experiment selects the most likely potentially associated entity pairs in the prediction results among the unknown entity associations as the output,and queries public medical databases to determine whether there is a real association.The top-10 case study on three diseases,including colon cancer,lymphoma,and lung cancer,all achieved 100% accuracy.