Predicting microRNA-disease associations from lncRNA-microRNA interactions via Multiview Multitask Learning.

Area of Science:

• Computational biology
• Genomics
• Biomedical informatics

Background:

• Identifying microRNA-disease associations (MDAs) is crucial for biomarker discovery.
• Existing computational methods rely on similarity assumptions that are not always valid.
• Long noncoding RNAs (lncRNAs) and microRNAs share regulatory relationships relevant to diseases.

Purpose of the Study:

• To propose a novel multiview multitask method for large-scale prediction of MDAs.
• To leverage lncRNA-microRNA interactions for MDA prediction, overcoming limitations of existing methods.
• To develop a deep learning model, MVMTMDA, for comprehensive microRNA representation.

Main Methods:

• Developed MVMTMDA, a deep learning model utilizing a multiview representation of microRNAs.
• Employed an end-to-end multitasking approach for training, enabling automatic determination of missing side information.
• Utilized known lncRNA-microRNA interactions as a basis for prediction.

Main Results:

• The MVMTMDA model achieved high performance with an average area under the ROC curve ranging from 0.8410 to 0.8521.
• The model effectively predicts MDAs even with incomplete microRNA information and without explicit similarity measurements.
• A statistical approach was proposed for predicting lncRNA-disease associations based on discovered MDAs.

Conclusions:

• The proposed MVMTMDA method offers a powerful new approach for discovering microRNA-disease associations.
• This work represents the first attempt to predict MDAs based on lncRNA-microRNA interactions.
• The findings have significant implications for identifying disease biomarkers and understanding gene regulation.