miTDS: Uncovering miRNA-mRNA interactions with deep learning for functional target prediction

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• MicroRNAs (miRNAs) regulate gene expression and are implicated in cancer.
• Identifying miRNA-mRNA interactions is crucial but challenging due to incomplete data and focus on known interactions.

Purpose of the Study:

• To develop a deep learning model, miTDS, for accurate identification of miRNA functional targets.
• To improve the prediction of unknown miRNA-mRNA interactions.
• To investigate aberrant RNA interactions, such as A-to-I editing, in cancer.

Main Methods:

• Developed miTDS, a deep learning model using dynamic word embeddings and transformer architecture.
• Incorporated a scoring mechanism to filter unstable sequence pairs.
• Utilized multi-scale attention and RNA-based dual-path modules for interaction analysis.

Main Results:

• miTDS demonstrated superior accuracy in predicting miRNA functional targets compared to state-of-the-art methods.
• The model was validated on datasets from MirTarbase and Diana-TarBase.
• miTDS successfully identified A-to-I RNA editing sites, offering insights into cancer suppression.

Conclusions:

• miTDS is an effective tool for predicting miRNA-mRNA functional targets.
• The model advances the understanding of miRNA-mRNA interactions and their role in cancer.
• miTDS has potential applications in identifying aberrant RNA interactions for cancer research.