EPMDA: an expression-profile based computational model for microRNA-disease association prediction.

Yu-An Huang,Lu-Xuan Xiang,Li-Ping Li,Xiao-Fang Li,Zhu-Hong You,Zhi-An Huang,Lin-Tao Lv

doi:10.18632/oncotarget.18788

Abstract

MicroRNA has become a new star molecule for understanding multiple biological processes and the mechanism of various complex human diseases. Even though a number of computational models have been proposed for predicting the association between microRNAs and various human diseases, most of them are mainly based on microRNA functional similarity and heterogeneous biological networks which suffer from inevitable computational error and bias. In this work, considering the limitation of information resource used by existing methods, we proposed EPMDA model which is the first computational method using the expression profiles of microRNAs to predict the most potential microRNAs associated with various diseases. Based on the dataset constructed from HMDD v2.0 database, EPMDA obtained AUCs of 0.8945 and 0.8917 based on the leave-one-out and 5-fold cross validation, respectively. Furthermore, EPMDA was applied to two important human diseases. As a result, 80% and 88% microRNAs in the top-25 lists of Colon Neoplasms and Kidney Neoplasms were confirmed by other databases. The performance comparison of EPMDA with existing prediction models and classical algorithms also demonstrated the reliable prediction ability of EPMDA. It is anticipated that EPMDA can be used as an effective computational tool for future biomedical researches.

Highlights

MicroRNAs are a kind of small non-coding RNA molecule which contains ~22 nucleotides and can be found in plants, animals and some viruses [1, 2]
In this work, considering the limitation of information resource used by existing methods, we proposed EPMDA model which is the first computational method using the expression profiles of microRNAs to predict the most potential microRNAs associated with various diseases
To evaluate the performance of EPMDA model, we implemented the method of leave-oneout cross validation (LOOCV) to predict the microRNA-disease association possibility based on the dataset downloaded from HMDD v2.0 database

Summary

Introduction

MicroRNAs (abbreviated miRNAs) are a kind of small non-coding RNA molecule which contains ~22 nucleotides and can be found in plants, animals and some viruses [1, 2]. As a breakthrough medical discovery, microRNA has been found to get involved in various biological processes [3, 4]. It can cause degradation and repression of RNA transcripts through complete or partial sequence complementarity, and further negatively regulates gene expression at the levels of messenger RNAs (mRNAs) [5, 6]. MicroRNAs, along with other kinds of noncoding RNA (i.e. lncRNA, circularRNA and snoRNA), has been considered as an important supplement for higher level of complexity and subtlety in human gene function

Methods

Results

Conclusion