MiRInter-Trans: a transformer-based framework for microRNA interaction prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Accurate microRNA (miRNA) interaction prediction is crucial for understanding gene regulation and developing RNA therapeutics.
• Current methods face challenges in capturing complex sequence patterns and predicting novel interactions.

Purpose of the Study:

• To introduce miRInter-Trans, a novel computational framework for predicting miRNA interactions.
• To leverage RNA foundation models and neural networks for sequence-based interaction prediction.
• To demonstrate the framework's capability in predicting de novo miRNA interactions.

Main Methods:

• Developed miRInter-Trans by combining a pre-trained RNA foundation model (RNA-FM) with a feed-forward neural network.
• Utilized transformer-based embeddings to capture sequence patterns and structural motifs.
• Did not rely on handcrafted features or thermodynamic parameters.

Main Results:

• miRInter-Trans achieved an Area Under the Receiver Operating Characteristic curve (AUROC) above 0.9 across multiple miRNA interaction types (miRNA-lncRNA, miRNA-miRNA, miRNA-snoRNA).
• Outperformed traditional Minimum Free Energy methods and other recent computational approaches.
• Demonstrated accurate de novo prediction capabilities for interactions with limited available data.

Conclusions:

• miRInter-Trans offers a powerful and accurate approach for miRNA interaction prediction solely from sequence data.
• The framework shows significant potential for advancing research in gene regulatory networks and RNA-based therapeutics.
• The developed model and datasets are publicly available for further research and application.