Predicting RNA-binding sites of proteins using support vector machines and evolutionary information

Area of Science:

• Computational biology
• Bioinformatics
• Molecular biology

Background:

• RNA-protein interactions are vital for numerous biological processes, including gene expression and viral infectivity.
• Experimental identification of RNA-binding sites is laborious; computational prediction is highly desirable.
• Existing computational methods often achieve high specificity at the cost of low sensitivity.

Purpose of the Study:

• To develop an accurate computational method for predicting RNA-binding sites in proteins.
• To improve the sensitivity of RNA-binding site prediction compared to existing methods.

Main Methods:

• Proposed RNAProB method incorporating a novel smoothed position-specific scoring matrix (PSSM) encoding scheme.
• Utilized a support vector machine (SVM) model for prediction.
• Employed a three-way data split procedure to prevent overfitting and ensure robust performance estimation.

Main Results:

• The smoothed PSSM encoding scheme significantly enhanced prediction performance, particularly sensitivity.
• RNAProB outperformed state-of-the-art systems in accuracy, specificity, and Matthew's correlation coefficient.
• Achieved substantial improvements in sensitivity (7.0%-26.9%) over benchmark datasets.

Conclusions:

• Smoothed PSSM encoding effectively improves RNA-binding site prediction by modeling residue dependencies.
• The method offers a more accurate way to distinguish between interacting and non-interacting residues.
• RNAProB has potential applications in predicting DNA-binding sites, protein-protein interactions, and posttranslational modification sites.