Since the microbiome has a significant impact on human health and disease, microbe-disease associations can be utilized as a valuable resource for understanding disease pathogenesis and promoting disease diagnosis and prognosis. Accordingly, it is necessary for researchers to achieve a comprehensive and deep understanding of the associations between microbes and diseases. Nevertheless, to date, little work has been achieved in implementing novel human microbe-disease association prediction models. In this paper, we develop a novel computational model to predict potential microbe-disease associations by bi-random walk on the heterogeneous network (BiRWHMDA). The heterogeneous network was constructed by connecting the microbe similarity network and the disease similarity network via known microbe-disease associations. Microbe similarity and disease similarity were calculated by the Gaussian interaction profile kernel similarity measure; moreover, a logistic function was applied to regulate disease similarity. Additionally, leave-one-out cross validation and 5-fold cross validation were implemented to evaluate the predictive performance of our method; both cross validation methods performed well. The leave-one-out cross validation experiment results illustrate that our method outperforms other previously proposed methods. Furthermore, case studies on asthma and inflammatory bowel disease prove the favorable performance of our method. In conclusion, our method can be considered as an effective computational model for predicting novel microbe-disease associations.

Increasing evidence shows that microbes are closely related to various human diseases. Obtaining a comprehensive and detailed understanding of the relationships between microbes and diseases would not only be beneficial to disease prevention, diagnosis and prognosis, but also would lead to the discovery of new drugs. However, because of a lack of data, little effort has been made to predict novel microbe-disease associations. To date, few methods have been proposed to solve the problem. In this study, we developed a new computational model based on network consistency projection to infer novel human microbe-disease associations (NCPHMDA) by integrating Gaussian interaction profile kernel similarity of microbes and diseases, and symptom-based disease similarity. NCPHMDA is a non-parametric and global network based model that combines microbe space projection and disease space projection to achieve the final prediction. Experimental results demonstrated that the integrated space projection of microbes and diseases, and symptom-based disease similarity played roles in the model performance. Cross validation frameworks and case studies further illustrated the superior predictive performance over other methods.

BackgroundWith the development of sequencing technology, more and more long non-coding RNAs (lncRNAs) have been identified. Some lncRNAs have been confirmed that they play an important role in the process of development through the dosage compensation effect, epigenetic regulation, cell differentiation regulation and other aspects. However, the majority of the lncRNAs have not been functionally characterized. Explore the function of lncRNAs and the regulatory network has become a hot research topic currently.MethodsIn the work, a network-based model named BiRWLGO is developed. The ultimate goal is to predict the probable functions for lncRNAs at large scale. The new model starts with building a global network composed of three networks: lncRNA similarity network, lncRNA-protein association network and protein-protein interaction (PPI) network. After that, it utilizes bi-random walk algorithm to explore the similarities between lncRNAs and proteins. Finally, we can annotate an lncRNA with the Gene Ontology (GO) terms according to its neighboring proteins.ResultsWe compare the performance of BiRWLGO with the state-of-the-art models on a manually annotated lncRNA benchmark with known GO terms. The experimental results assert that BiRWLGO outperforms other methods in terms of both maximum F-measure (Fmax) and coverage.ConclusionsBiRWLGO is a relatively efficient method to predict the functions of lncRNA. When protein interaction data is integrated, the predictive performance of BiRWLGO gains a great improvement.Electronic supplementary materialThe online version of this article (10.1186/s12920-018-0414-2) contains supplementary material, which is available to authorized users.

No abstract

No abstract