Related Experiment Video
Updated: Sep 6, 2025
06:16
mirMachine: A One-Stop Shop for Plant miRNA Annotation
Published on: May 1, 2021
2.6K
A deep learning method for miRNA/isomiR target detection.
Amlan Talukder1, Wencai Zhang2, Xiaoman Li3
1Department of Computer Science, University of Central Florida, Orlando, FL, 32816, USA.
Scientific Reports
|June 23, 2022
Summary
We developed DMISO, a deep learning tool for microRNA (miRNA) target prediction that accounts for isomiRs. DMISO significantly improves accuracy by analyzing intricate miRNA/isomiR-mRNA interactions.
Area of Science:
- Computational Biology
- Bioinformatics
- Genomics
Background:
- Accurate microRNA (miRNA) target identification at base-pair resolution remains a challenge.
- The existence of miRNA isoforms (isomiRs) complicates existing prediction methods.
- Current miRNA target prediction tools often overlook isomiRs and exhibit suboptimal performance.
Purpose of the Study:
- To improve the accuracy of miRNA target predictions by incorporating isomiR-mRNA interactions.
- To develop a deep learning model capable of capturing complex features of miRNA/isomiR-mRNA interactions.
Main Methods:
- Development of a deep learning tool named DMISO.
- DMISO analyzes intricate features of miRNA/isomiR-mRNA interactions.
- Evaluation using tenfold cross-validation and three independent datasets.
Main Results:
- DMISO achieved high precision (95%) and recall (90%) in tenfold cross-validation.
- DMISO outperformed five other tools, including conventional and deep learning-based methods, on independent datasets.
- Feature interpretation highlighted the importance of non-seed miRNA regions and RNA-binding motifs.
Conclusions:
- DMISO offers a significant advancement in miRNA target prediction by integrating isomiR analysis.
- The model demonstrates superior performance compared to existing state-of-the-art tools.
- Understanding non-seed regions and RNA-binding motifs is crucial for accurate miRNA-mRNA interaction prediction.
