A Novel Computational Method for the Identification of Potential miRNA-Disease Association Based on Symmetric Non-negative Matrix Factorization and Kronecker Regularized Least Square.

Yan Zhao,Xing Chen,Jun Yin

doi:10.3389/fgene.2018.00324

Yan Zhao, Xing Chen + Show 1 more

Open Access

https://doi.org/10.3389/fgene.2018.00324

Copy DOI

Abstract

Increasing evidence has indicated that microRNAs (miRNAs) are associated with numerous human diseases. Studying the associations between miRNAs and diseases contributes to the exploration of effective diagnostic and treatment approaches for diseases. Unfortunately, the use of biological experiments to reveal the potential associations between miRNAs and diseases is time consuming and costly. Therefore, it is very necessary to use simple and efficient calculation models to predict potential disease-related miRNAs. Considering the limitations of other previous methods, we proposed a novel computational model of Symmetric Nonnegative Matrix Factorization for MiRNA-Disease Association prediction (SNMFMDA) to reveal the relation of miRNA-disease pairs. SNMFMDA could be applied to predict miRNAs associated with new diseases. Compared to the direct use of the integrated similarity in previous computational models, the integrated similarity need to be interpolated by symmetric non-negative matrix factorization (SymNMF) before application in SNMFMDA, and the relevant probability of disease-miRNA was obtained mainly through Kronecker regularized least square (KronRLS) method in our model. What's more, the AUC of global leave-one-out cross validation (LOOCV) reached 0.9007, and the AUC based on local LOOCV was 0.8426. Besides, the mean and the standard deviation of AUCs achieved 0.8830 and 0.0017 respectively in 5-fold cross validation. All of the above results demonstrated the superior prediction performance of SNMFMDA. We also conducted three different case studies on Esophageal Neoplasms, Breast Neoplasms and Lung Neoplasms, and 49, 49, and 48 of the top 50 of their predicted miRNAs respectively were confirmed by databases or related literatures. It could be expected that SNMFMDA would be a model with the ability to predict disease-related miRNAs efficiently and accurately.

Highlights

MicroRNAs are a class of endogenous non-coding RNAs with regulatory functions found in eukaryotes, which are approximately 20–25 nucleotides in length (Ambros, 2001)
In this study, based on the 5,430 confirmed associations between 383 diseases and 495 miRNAs recorded in HMDD v2.0, local and global leave-one-out cross validation (LOOCV) were applied to test the prediction performance of SNMFMDA
In LOOCV, each of the 5,430 known associations was left out in turn as the test sample and the remaining 5,429 known associations were considered as training samples, while the miRNA-disease pairs without verified association were viewed as candidate samples, and we applied SNMFMDA to calculated the association probability of candidate samples and test sample

Summary

Introduction

MicroRNAs (miRNAs) are a class of endogenous non-coding RNAs with regulatory functions found in eukaryotes, which are approximately 20–25 nucleotides in length (Ambros, 2001). It has been verified that miRNAs are crucial constituent in cells and may make an important impact in many important biological processes, including proliferation (Cheng et al, 2005), development (Karp and Ambros, 2005), differentiation (Miska, 2005), viral infection (Miska, 2005) and so on. It is taken for granted that there are associations between miRNAs and the generation as well as development of a number of human diseases (Alvarez-Garcia and Miska, 2005). Since a large amount of biological data has been collected and organized into different databases, it is feasible and necessary to develop computational model to reveal potential disease-miRNA associations based on these databases (Chen, 2015)

Methods

Results

Conclusion