A systematic benchmark of machine learning methods for protein-RNA interaction prediction.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA-binding proteins (RBPs) regulate gene expression post-transcriptionally.
• In vivo RBP-binding site identification is limited by transcript availability.
• Computational methods are needed to infer missing RBP-binding information.

Purpose of the Study:

• To benchmark machine learning methods for in vivo RNA-binding protein-RNA interaction prediction.
• To provide a standardized framework for evaluating prediction methods.
• To guide researchers in selecting optimal prediction tools and aid developers in creating new methods.

Main Methods:

• Compiled 37 machine learning methods for RBP-RNA interaction prediction.
• Systematically benchmarked 11 representative methods using hundreds of CLIP-seq datasets.
• Employed homogenized sample pre-processing and two negative-class sample generation strategies.

Main Results:

• Evaluated predictive performance of benchmarked methods.
• Assessed the impact of neural network architectures and input modalities.
• Identified key factors influencing model performance in RBP-binding site prediction.

Conclusions:

• The study provides a comprehensive comparison of computational methods for RBP-RNA interaction prediction.
• A standardized evaluation framework is introduced to facilitate method selection and development.
• Findings will empower researchers to choose appropriate tools and guide future advancements in the field.