beRBP: binding estimation for human RNA-binding proteins.

Area of Science:

• Computational biology
• Molecular biology
• Bioinformatics

Background:

• Understanding RNA-binding protein (RBP) interactions is crucial for elucidating their functional mechanisms.
• Current computational methods for predicting RBP-RNA interactions often involve machine learning, facing a trade-off between prediction accuracy (RBP-specific models) and broad applicability (pooled RBP-RNA interaction models).
• RBP-specific models require substantial known targets, limiting their use, while pooled models obscure crucial interaction features, reducing accuracy.

Purpose of the Study:

• To present beRBP, a novel dual computational approach for predicting human RBP-RNA interactions.
• To provide a tool that can accurately predict interactions for both RBPs with abundant known targets and those with limited or no known targets.
• To improve the scope and accuracy of computational RBP binding estimation.

Main Methods:

• beRBP utilizes a dual approach based on Random Forests.
• It builds RBP-specific models for proteins with sufficient known targets.
• It develops a general model for RBPs with limited or null known targets, using PWM of a RBP and an RNA sequence.

Main Results:

• Both the specific and general models of beRBP demonstrated strong performance compared to existing methods on benchmark datasets.
• The general model of beRBP outperformed existing methods on most novel RBPs.
• beRBP offers a composite solution that integrates RBP-specific and general prediction strategies.

Conclusions:

• beRBP is a promising computational tool for estimating human RBP binding.
• The dual approach of beRBP provides good prediction accuracy and a broad application scope.
• This tool facilitates a deeper understanding of RBP functional mechanisms through accurate binding site prediction.