The linear neighborhood propagation method for predicting long non-coding RNA - protein interactions

Long non-coding RNAs (lncRNAs) have gained wide attentions because of their essential functions in a variety of biological processes. Though precise functions and mechanisms of most lncRNAs remain unknown, studies show that lncRNAs generally exert functions through interactions with the corresponding RNA-binding proteins. The experimental detection of lncRNA-protein interactions is costly and time-consuming. In this paper, we propose a linear neighborhood propagation method (LPLNP), to predict lncRNA-protein interactions. LPLNP calculates the linear neighborhood similarity in the feature space, and transfers it into the interaction space, and predict unobserved interactions between the lncRNAs and proteins by a label propagation process. Our study shows that the LPLNP model based on the known lncRNA-protein interactions can produce high-accuracy performances, achieving an AUPR score of 0.42. Furthermore, we incorporate biological information of lncRNAs and proteins into the LPLNP model, and can further increase the performances, achieving an AUPR score of 0.4584. The case study demonstrates that many lncRNA-protein interactions predicted by our method can be validated, indicating that our method is a useful tool for lncRNA-protein interaction prediction. The source code and the dataset used in the paper are available at: https://github.com/BioMedicalBigDataMiningLabWhu/lncRNA-protein-interaction-prediction. (C) 2017 Elsevier B.V. All rights reserved.