miRAW: A deep learning-based approach to predict microRNA targets by analyzing whole microRNA transcripts

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) regulate gene expression post-transcriptionally.
• Current miRNA target prediction methods often rely on the seed region, missing many functional interactions.
• A more comprehensive approach is needed to capture the full complexity of miRNA-mRNA binding.

Purpose of the Study:

• To develop a novel Deep Learning (DL) model for accurate miRNA target site prediction.
• To investigate the role of the entire miRNA and mRNA sequence in targeting, beyond the traditional seed region.
• To outperform existing computational methods in identifying functional miRNA:gene targets.

Main Methods:

• Collected over 150,000 validated human miRNA:gene targets.
• Integrated CLIP-Seq, CLASH, and iPAR-CLIP data to identify ~20,000 exact target sites.
• Implemented a deep neural network with autoencoders and a feed-forward network for feature learning.
• Incorporated site location and accessibility energy for prediction refinement.

Main Results:

• The DL model automatically learned features describing miRNA-mRNA interactions.
• The approach identified the seed region's importance and the role of pairings outside this region.
• Performance comparison on independent datasets showed the DL method consistently outperformed existing prediction tools.
• Thermodynamic analysis indicated site accessibility is a factor but not solely indicative of functionality.

Conclusions:

• Deep Learning offers a powerful, flexible methodology for miRNA target prediction.
• Considering the entire miRNA-mRNA interaction provides a more accurate understanding of gene regulation.
• The developed DL approach advances the field of miRNA target identification and functional analysis.