Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network Regularization
The identification of potential microRNA (miRNA)-disease associations enables the elucidation of the pathogenesis of complex human diseases owing to the crucial role of miRNAs in various biologic processes and it yields insights into novel prognostic markers. In the consideration of the time and cos...
Main Authors: | , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2020-04-01
|
Series: | Cells |
Subjects: | |
Online Access: | https://www.mdpi.com/2073-4409/9/4/881 |
id |
doaj-377d828f64f64b1eac4d158f83544a30 |
---|---|
record_format |
Article |
spelling |
doaj-377d828f64f64b1eac4d158f83544a302020-11-25T02:23:52ZengMDPI AGCells2073-44092020-04-01988188110.3390/cells9040881Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network RegularizationJihwan Ha0Chihyun Park1Chanyoung Park2Sanghyun Park3Department of Computer Science, Yonsei University, Seoul 03722, KoreaDepartment of Computer Science, Yonsei University, Seoul 03722, KoreaDepartment of Computer Science, University of Illinois at Urbana-Champaign, Urbana, OH 61801, USADepartment of Computer Science, Yonsei University, Seoul 03722, KoreaThe identification of potential microRNA (miRNA)-disease associations enables the elucidation of the pathogenesis of complex human diseases owing to the crucial role of miRNAs in various biologic processes and it yields insights into novel prognostic markers. In the consideration of the time and costs involved in wet experiments, computational models for finding novel miRNA-disease associations would be a great alternative. However, computational models, to date, are biased towards known miRNA-disease associations; this is not suitable for rare miRNAs (i.e., miRNAs with a few known disease associations) and uncommon diseases (i.e., diseases with a few known miRNA associations). This leads to poor prediction accuracies. The most straightforward way of improving the performance is by increasing the number of known miRNA-disease associations. However, due to lack of information, increasing attention has been paid to developing computational models that can handle insufficient data via a technical approach. In this paper, we present a general framework—improved prediction of miRNA-disease associations (IMDN)—based on matrix completion with network regularization to discover potential disease-related miRNAs. The success of adopting matrix factorization is demonstrated by its excellent performance in recommender systems. This approach considers a miRNA network as additional implicit feedback and makes predictions for disease associations relevant to a given miRNA based on its direct neighbors. Our experimental results demonstrate that IMDN achieved excellent performance with reliable area under the receiver operating characteristic (ROC) area under the curve (AUC) values of 0.9162 and 0.8965 in the frameworks of global and local leave-one-out cross-validations (LOOCV), respectively. Further, case studies demonstrated that our method can not only validate true miRNA-disease associations but also suggest novel disease-related miRNA candidates.https://www.mdpi.com/2073-4409/9/4/881miRNAdiseasemiRNA-disease associationmiRNA similarity networkmatrix factorization |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Jihwan Ha Chihyun Park Chanyoung Park Sanghyun Park |
spellingShingle |
Jihwan Ha Chihyun Park Chanyoung Park Sanghyun Park Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network Regularization Cells miRNA disease miRNA-disease association miRNA similarity network matrix factorization |
author_facet |
Jihwan Ha Chihyun Park Chanyoung Park Sanghyun Park |
author_sort |
Jihwan Ha |
title |
Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network Regularization |
title_short |
Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network Regularization |
title_full |
Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network Regularization |
title_fullStr |
Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network Regularization |
title_full_unstemmed |
Improved Prediction of miRNA-Disease Associations Based on Matrix Completion with Network Regularization |
title_sort |
improved prediction of mirna-disease associations based on matrix completion with network regularization |
publisher |
MDPI AG |
series |
Cells |
issn |
2073-4409 |
publishDate |
2020-04-01 |
description |
The identification of potential microRNA (miRNA)-disease associations enables the elucidation of the pathogenesis of complex human diseases owing to the crucial role of miRNAs in various biologic processes and it yields insights into novel prognostic markers. In the consideration of the time and costs involved in wet experiments, computational models for finding novel miRNA-disease associations would be a great alternative. However, computational models, to date, are biased towards known miRNA-disease associations; this is not suitable for rare miRNAs (i.e., miRNAs with a few known disease associations) and uncommon diseases (i.e., diseases with a few known miRNA associations). This leads to poor prediction accuracies. The most straightforward way of improving the performance is by increasing the number of known miRNA-disease associations. However, due to lack of information, increasing attention has been paid to developing computational models that can handle insufficient data via a technical approach. In this paper, we present a general framework—improved prediction of miRNA-disease associations (IMDN)—based on matrix completion with network regularization to discover potential disease-related miRNAs. The success of adopting matrix factorization is demonstrated by its excellent performance in recommender systems. This approach considers a miRNA network as additional implicit feedback and makes predictions for disease associations relevant to a given miRNA based on its direct neighbors. Our experimental results demonstrate that IMDN achieved excellent performance with reliable area under the receiver operating characteristic (ROC) area under the curve (AUC) values of 0.9162 and 0.8965 in the frameworks of global and local leave-one-out cross-validations (LOOCV), respectively. Further, case studies demonstrated that our method can not only validate true miRNA-disease associations but also suggest novel disease-related miRNA candidates. |
topic |
miRNA disease miRNA-disease association miRNA similarity network matrix factorization |
url |
https://www.mdpi.com/2073-4409/9/4/881 |
work_keys_str_mv |
AT jihwanha improvedpredictionofmirnadiseaseassociationsbasedonmatrixcompletionwithnetworkregularization AT chihyunpark improvedpredictionofmirnadiseaseassociationsbasedonmatrixcompletionwithnetworkregularization AT chanyoungpark improvedpredictionofmirnadiseaseassociationsbasedonmatrixcompletionwithnetworkregularization AT sanghyunpark improvedpredictionofmirnadiseaseassociationsbasedonmatrixcompletionwithnetworkregularization |
_version_ |
1724856760059035648 |