Artificial intelligence methods enhance the discovery of RNA interactions

Area of Science:

• Computational biology
• Bioinformatics
• Molecular biology

Background:

• RNA interactions with proteins and other RNAs are crucial in cellular processes.
• Machine learning (ML) and deep learning (DL) methods are increasingly used to predict these interactions.
• Current prediction methods often lack standardized evaluation across diverse datasets.

Purpose of the Study:

• To review and compare recent computational methods for predicting RNA-protein and RNA-RNA interactions.
• To analyze feature encoding and machine learning strategies employed in these methods.
• To discuss the impact of dataset characteristics and negative sampling on prediction performance.

Main Methods:

• Systematic review of computational methods for RNA interaction prediction.
• Analysis of feature encoding techniques and ML/DL algorithms.
• Evaluation of dataset properties and negative data generation strategies.

Main Results:

• Identified various feature encoding and ML strategies for RNA interaction prediction.
• Highlighted the critical role of dataset size and quality in model performance.
• Summarized best-performing methods for predicting interactions involving specific RNA types (circRNAs, lncRNAs) and general RNA-RNA/RNA-protein interactions.

Conclusions:

• Standardized evaluation across datasets is needed for robust comparison of RNA interaction prediction methods.
• Dataset characteristics and appropriate negative sampling are key for developing accurate predictive models.
• The review provides insights into state-of-the-art computational approaches for understanding RNA-based molecular interactions.