Advancing microRNA target site prediction with transformer and base-pairing patterns

Area of Science:

• Molecular Biology
• Bioinformatics
• Genetics

Background:

• MicroRNAs (miRNAs) are key regulators of gene expression, crucial for cellular processes.
• Identifying miRNA targets is essential for understanding complex gene regulatory networks.
• Traditional methods focus on canonical miRNA-target interactions, potentially missing regulatory mechanisms.

Purpose of the Study:

• To develop a novel computational approach, Mimosa, for enhanced prediction of miRNA targets.
• To improve the identification of both canonical and non-canonical miRNA binding sites.
• To provide a user-friendly web server for miRNA target prediction.

Main Methods:

• Developed Mimosa, a computational approach utilizing the Transformer framework.
• Integrated contextual, positional, and base-pairing information for improved prediction accuracy.
• Benchmarked Mimosa's performance on gene-level and site-level predictions across multiple species.

Main Results:

• Mimosa demonstrates superior performance in gene-level miRNA target predictions.
• The model shows impressive accuracy in predicting miRNA binding sites, including non-canonical interactions.
• Extensive benchmarking confirms Mimosa's effectiveness across various non-human species.

Conclusions:

• Mimosa offers a significant advancement in miRNA target prediction by effectively identifying non-canonical sites.
• The tool reduces reliance on pre-selected candidate targets, offering a more comprehensive analysis.
• A publicly available web server facilitates broader research application of Mimosa.