Efficient framework for predicting MiRNA-disease associations based on improved hybrid collaborative filtering

BMC Medical Informatics and Decision Making

Ru Nie¹  Jiashu Li¹  Zhengwei Li²  Zhu-hong You³  Wenzheng Bao⁴ 

[1] Engineering Research Center of Mine Digitalization of Ministry of Education, China University of Mining and Technology, 221116, Xuzhou, China;School of Computer Science and Technology, China University of Mining and Technology, 221116, Xuzhou, China;Engineering Research Center of Mine Digitalization of Ministry of Education, China University of Mining and Technology, 221116, Xuzhou, China;School of Computer Science and Technology, China University of Mining and Technology, 221116, Xuzhou, China;Institute of Machine Learning and Systems Biology, College of Electronics and Information Engineering, Tongji University, 201804, Shanghai, China;KUNPAND Communications (Kunshan) Co., Ltd., 215300, Suzhou, China;School of Computer Science, Northwestern Polytechnical University, 710072, Xi’an, China;School of Information Engineering, Xuzhou University of Technology, 221018, Xuzhou, China;

关键词: miRNA-disease association prediction;    Hybrid collaborative filtering;    Heterogeneous data;    Singular value decomposition;   

DOI  :  10.1186/s12911-021-01616-5

来源: Springer

【 摘 要 】

BackgroundAccumulating studies indicates that microRNAs (miRNAs) play vital roles in the process of development and progression of many human complex diseases. However, traditional biochemical experimental methods for identifying disease-related miRNAs cost large amount of time, manpower, material and financial resources.MethodsIn this study, we developed a framework named hybrid collaborative filtering for miRNA-disease association prediction (HCFMDA) by integrating heterogeneous data, e.g., miRNA functional similarity, disease semantic similarity, known miRNA-disease association networks, and Gaussian kernel similarity of miRNAs and diseases. To capture the intrinsic interaction patterns embedded in the sparse association matrix, we prioritized the predictive score by fusing three types of information: similar disease associations, similar miRNA associations, and similar disease-miRNA associations. Meanwhile, singular value decomposition was adopted to reduce the impact of noise and accelerate predictive speed.ResultsWe then validated HCFMDA with leave-one-out cross-validation (LOOCV) and two types of case studies. In the LOOCV, we achieved 0.8379 of AUC (area under the curve). To evaluate the performance of HCFMDA on real diseases, we further implemented the first type of case validation over three important human diseases: Colon Neoplasms, Esophageal Neoplasms and Prostate Neoplasms. As a result, 44, 46 and 44 out of the top 50 predicted disease-related miRNAs were confirmed by experimental evidence. Moreover, the second type of case validation on Breast Neoplasms indicates that HCFMDA could also be applied to predict potential miRNAs towards those diseases without any known associated miRNA.ConclusionsThe satisfactory prediction performance demonstrates that our model could serve as a reliable tool to guide the following research for identifying candidate miRNAs associated with human diseases.

【 授权许可】

CC BY