econvRBP: Improved ensemble convolutional neural networks for RNA binding protein prediction directly from sequence

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA binding proteins (RBPs) are crucial for RNA metabolism, influencing transport, translation, and degradation.
• Accurate identification of RBPs is vital for genome annotation, but computational prediction faces challenges in feature extraction and integration.

Purpose of the Study:

• To develop a novel computational method for predicting RNA binding proteins (RBPs) using amino acid sequences.
• To address limitations in existing methods, including shallow feature representation and poor feature fusion.

Main Methods:

• A novel ensemble convolutional neural network (econvRBP) was developed.
• One Hot and Conjoint Triad encoding methods were used to capture local and global sequence features.
• Features were fused and processed using convolutional neural networks for high-level feature extraction.

Main Results:

• The econvRBP model achieved 99% accuracy in predicting 2875 RBPs and 6782 non-RBPs.
• The method demonstrated superior performance compared to existing predictors in 10-fold cross-validation.
• Validation on RBPPred datasets yielded an accuracy of 0.87.

Conclusions:

• The econvRBP method represents a significant advancement in predicting RNA binding proteins.
• This approach offers reliable guidance for RBP detection and enhances genome annotation.
• The econvRBP tool is publicly available for further research and application.