DANE-MDA: Predicting microRNA-disease associations via deep attributed network embedding

Yi Wang,Zheng-Wei Li,Zhu-Hong You,Leon Wong,Bo-Ya Ji

doi:10.1016/j.isci.2021.102455

Abstract

SummaryPredicting the microRNA-disease associations by using computational methods is conductive to the efficiency of costly and laborious traditional bio-experiments. In this study, we propose a computational machine learning-based method (DANE-MDA) that preserves integrated structure and attribute features via deep attributed network embedding to predict potential miRNA-disease associations. Specifically, the integrated features are extracted by using deep stacked auto-encoder on the diverse orders of matrixes containing structure and attribute information and are then trained by using random forest classifier. Under 5-fold cross-validation experiments, DANE-MDA yielded average accuracy, sensitivity, and AUC at 85.59%, 84.23%, and 0.9264 in term of HMDD v3.0 dataset, and 83.21%, 80.39%, and 0.9113 in term of HMDD v2.0 dataset, respectively. Additionally, case studies on breast, colon, and lung neoplasms related disease show that 47, 47, and 46 of the top 50 miRNAs can be predicted and retrieved in the other database.

Highlights

The human genomes have various endogenous ‘‘non-messenger’’ or ‘‘non-coding’’ RNAs, including a large number of single-stranded microRNAs containing about 22 nucleotides (Ambros, 2001, 2004). miRNAs play a significant function in various human life processes, including virus defense, tissue development, cell metabolism, and organ formation, and participate in the regulation of post-transcriptional gene expression (Cui et al, 2006; Karp and Ambros, 2005; Lu et al, 2005; Rupaimoole and Slack, 2017; Xu et al, 2004)
SUMMARY Predicting the microRNA-disease associations by using computational methods is conductive to the efficiency of costly and laborious traditional bio-experiments
We propose a computational machine learning-based method (DANE-MDA) that preserves integrated structure and attribute features via deep attributed network embedding to predict potential miRNA-disease associations

Summary

Introduction

The human genomes have various endogenous ‘‘non-messenger’’ or ‘‘non-coding’’ RNAs, including a large number of single-stranded microRNAs (miRNAs) containing about 22 nucleotides (Ambros, 2001, 2004). miRNAs play a significant function in various human life processes, including virus defense, tissue development, cell metabolism, and organ formation, and participate in the regulation of post-transcriptional gene expression (Cui et al, 2006; Karp and Ambros, 2005; Lu et al, 2005; Rupaimoole and Slack, 2017; Xu et al, 2004). MiRNAs play a significant function in various human life processes, including virus defense, tissue development, cell metabolism, and organ formation, and participate in the regulation of post-transcriptional gene expression (Cui et al, 2006; Karp and Ambros, 2005; Lu et al, 2005; Rupaimoole and Slack, 2017; Xu et al, 2004). Massive miRNA-disease associations have been acquired through traditional biological experiments and stored in public databases. These biological experimental methods usually have high prediction accuracy; their processes are complex, expensive, and time-consuming (Liang et al, 2019). To accelerate the verification process, and reduce the time consumption and blindness of biological experiments, it is significant to establish computational methods for quickly and effectively predicting possible miRNA-disease associations (Wong et al, 2020; Yi et al, 2020)

Methods

Results

Discussion

Conclusion