Journal
BRIEFINGS IN BIOINFORMATICS
Volume 22, Issue 6, Pages -Publisher
OXFORD UNIV PRESS
DOI: 10.1093/bib/bbab328
Keywords
microRNA; small molecule; association prediction; matrix completion; bounded nuclear norm regularization
Funding
- National Natural Science Foundation of China [61972399, 11931008]
Ask authors/readers for more resources
The study developed a computational model for predicting small molecule-miRNA associations using Bounded Nuclear Norm Regularization method. The innovation lies in limiting matrix elements within (0,1) and incorporating a regularization term to tolerate noise.
Mounting evidence has demonstrated the significance of taking microRNAs (miRNAs) as the target of small molecule (SM) drugs for disease treatment. Given the fact that exploring new SM-miRNA associations through biological experiments is extremely expensive, several computing models have been constructed to reveal the possible SM-miRNA associations. Here, we built a computing model of Bounded Nuclear Norm Regularization for SM-miRNA Associations prediction (BNNRSMMA). Specifically, we first constructed a heterogeneous SM-miRNA network utilizing miRNA similarity, SM similarity, confirmed SM-miRNA associations and defined a matrix to represent the heterogeneous network. Then, we constructed a model to complete this matrix by minimizing its nuclear norm. The Alternating Direction Method of Multipliers was adopted to minimize the nuclear norm and obtain predicted scores. The main innovation lies in two aspects. During completion, we limited all elements of the matrix within the interval of (0,1) to make sure they have practical significance. Besides, instead of strictly fitting all known elements, a regularization term was incorporated to tolerate the noise in integrated similarities. Furthermore, four kinds of cross-validations on two datasets and two types of case studies were performed to evaluate the predictive performance of BNNRSMMA. Finally, BNNRSMMA attained areas under the curve of 0.9822 (0.8433), 0.9793 (0.8852), 0.8253 (0.7350) and 0.9758 +/- 0.0029 (0.8759 +/- 0.0041) under global leave-one-out cross-validation (LOOCV), miRNA-fixed LOOCV, SM-fixed LOOCV and 5-fold cross-validation based on Dataset 1(Dataset 2), respectively. With regard to case studies, plenty of predicted associations have been verified by experimental literatures. All these results confirmed that BNNRSMMA is a reliable tool for inferring associations.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available