From binary labels to dynamic landscapes: The evolving computational prediction of protein-RNA interactions through...

Area of Science:

• Computational biology
• Bioinformatics
• Molecular biology

Background:

• Protein-RNA interactions (RPIs) are crucial for post-transcriptional gene regulation.
• Computational approaches for studying RPIs have rapidly advanced.
• Existing methods address various aspects, including RBP classification and binding affinity prediction.

Purpose of the Study:

• To review the methodological evolution of computational RPI prediction from 2010-2025.
• To compare different computational approaches, including machine learning and deep learning.
• To highlight recent advances and provide recommendations for future research.

Main Methods:

• Systematic review of computational methods for RPI prediction.
• Classification of methods into five primary categories.
• Comparison of deep learning, graph neural networks, and large-scale pre-trained language models.
• Emphasis on structure-aware, condition-aware, and low-data regime learning.

Main Results:

• Computational RPI prediction has evolved from conventional machine learning to sophisticated deep learning models.
• Significant progress has been made in structure-aware and condition-aware modeling.
• Challenges remain in data preparation, evaluation protocols, and generalization behavior.

Conclusions:

• The field has seen rapid methodological advancements in computational RPI prediction.
• Future directions include integration with spatial omics and development of dynamic RPI models.
• Standardized benchmarking is recommended for robust field-wide progress.