Predicting protein-binding RNA nucleotides with consideration of binding partners

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• Predicting RNA-binding sites in proteins is established, but predicting protein-binding sites in RNA is challenging with lower accuracy.
• Existing methods often overlook the specific interacting protein partner, limiting predictions for proteins binding to various RNAs.
• Our prior work developed a method for predicting protein-binding nucleotides in RNA sequences.

Purpose of the Study:

• To enhance the accuracy and utility of predicting protein-binding nucleotides in RNA.
• To develop a novel computational model that integrates both RNA and protein sequence data for improved predictions.
• To identify key sequence-based features for RNA-protein binding site prediction.

Main Methods:

• Developed a new support vector machine (SVM) model incorporating features from both RNA and protein sequences.
• Identified and selected effective molecular features predictive of RNA-protein interactions.
• Validated the model using 10-fold cross-validation and independent testing datasets.

Main Results:

• The new model, utilizing both RNA and protein sequences, achieved a 10-fold cross-validation sensitivity of 86.5% and specificity of 86.2%.
• Independent testing of the integrated model yielded a sensitivity of 58.8% and specificity of 87.4%.
• The combined sequence model outperformed a model using only RNA sequence data across most performance metrics in both validation and testing phases.

Conclusions:

• This study presents the first sequence-based prediction of protein-binding nucleotides in RNA that accounts for the specific binding partner.
• The novel SVM model demonstrates superior performance compared to RNA-only sequence models.
• The findings offer valuable insights for experimental design aimed at discovering protein-binding sites in RNA molecules with unknown structures.