aPRBind

aPRBind predicts RNA-binding residues in proteins using a convolutional neural network that integrates sequence-based and structural features to identify protein–RNA interaction sites.

Key Features:

• Ab-initio Prediction: Performs ab-initio prediction of RNA-binding residues without requiring prior knowledge of RNA-binding sites.
• CNN Architecture: Uses a convolutional neural network to process and integrate input features for residue-level prediction.
• Sequence-based Features: Incorporates sequence features such as amino acid composition.
• Structural Features from I-TASSER: Utilizes structural features derived from protein structures predicted by I-TASSER, including residue dynamics.
• Residue-Nucleotide Propensity: Includes residue-nucleotide propensity as an explicit predictive feature.
• Complementary Feature Contribution: Sequence features provide the dominant predictive signal while structural features (e.g., dynamics) offer complementary information.
• Robustness to Model Accuracy: Maintains predictive utility on modeled structures with TM-score ≥ 0.5 and shows reduced sensitivity to structural model precision.

Scientific Applications:

• Protein-RNA Interaction Research: Enables identification of RNA-binding residues to support studies of protein–RNA interactions involved in gene regulation and viral replication.
• Modeling and Structural Biology: Assists modeling and analysis of proteins with unknown structures or refined models, including applications where modeled structures have TM-score ≥ 0.5.
• Drug Discovery and Design: Informs targeting of protein RNA-binding sites for therapeutic strategy development.

Methodology:

Trained using a combination of sequence features (e.g., amino acid composition) and structural features including residue dynamics and residue-nucleotide propensity extracted from I-TASSER-predicted structures, with a convolutional neural network integrating these features to predict RNA-binding residues.