MvKFN-MDA: Multi-view Kernel Fusion Network for miRNA-disease association prediction
Jin Li,Tao Liu,Jingru Wang,Qing Li,Chenxi Ning,Yun Yang
DOI: https://doi.org/10.1016/j.artmed.2021.102115
IF: 7.011
2021-08-01
Artificial Intelligence in Medicine
Abstract:<p>Predicting the associations between microRNAs (miRNAs) and diseases is of great significance for identifying miRNAs related to human diseases. Since it is time-consuming and costly to identify the association between miRNA and disease through biological experiments, computational methods are currently used as an effective supplement to identify the potential association between disease and miRNA. This paper presents a Multi-view Kernel Fusion Network (MvKFN) based prediction method (MvKFN-MDA) to address the problem of miRNA-disease associations prediction. A novel multiple kernel fusion framework Multi-view Kernel Fusion Network (MvKFN) is first proposed to effectively fuse different views similarity kernels constructed from different data sources in a highly nonlinear way. Using MvKFNs, both different base similarity kernels for miRNA, such as sequence, functional, semantic, Gaussian profile kernels and different base similarity kernels for diseases, such as semantic, Gaussian profile kernel are nonlinearly fused into two integrated similarity kernels, one for miRNA, another for disease. Then, miRNA and disease feature representations are extracted from the miRNA and disease integrated similarity kernels respectively. These features are then fed into a neural matrix completion framework which finally outputs the association prediction scores. The parameters of MvKFN-MDA are learned based on the known miRNA-disease association matrix in a supervised end-to-end way. We compare the proposed method with other state-of-the-art methods. The AUCs of our proposed method were superior to the existing methods in both 5-FCV and LOOCV on two open experimental datasets. Furthermore, 49, 48, and 47 of the top 50 predicted miRNAs for three high-risk human diseases, namely, colon cancer, lymphoma, and kidney cancer, are verified respectively using experimental literature. Finally, 100% accuracy from the top 50 predicted miRNAs is achieved when breast cancer is used as a case study to evaluate the ability of MvKFN-MDA for predicting a new disease without any known related miRNAs.</p>
engineering, biomedical,computer science, artificial intelligence,medical informatics