aPRBind: protein-RNA interface prediction by combining sequence and I-TASSER model-based structural features learned...

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Protein-RNA interactions are vital for numerous biological processes.
• Accurate prediction of RNA-binding residues in proteins is a significant challenge in computational biology.
• Current sequence-based ab-initio methods exhibit limited accuracy.

Purpose of the Study:

• To develop a novel ab-initio method, aPRBind, for enhanced RNA-binding residue prediction.
• To leverage convolutional neural networks and integrate diverse sequence and structural features.
• To improve the accuracy and applicability of RNA-binding site prediction.

Main Methods:

• Developed aPRBind, a convolutional neural network-based ab-initio method.
• Utilized sequence features and structural information, including residue dynamics and residue-nucleotide propensity.
• Extracted features from protein structures predicted by I-TASSER.

Main Results:

• aPRBind demonstrates superior performance compared to state-of-the-art ab-initio methods on benchmark datasets.
• Sequence features were identified as the most critical, followed by dynamics information.
• Structural features are complementary to sequence features, enhancing binding site prediction accuracy.
• The method shows robustness with modeled structures (TM-score≥0.5) and has marginal dependence on structure model accuracy.

Conclusions:

• aPRBind offers a significant advancement in ab-initio RNA-binding residue prediction.
• The integration of sequence and structural features, particularly dynamics, improves prediction accuracy.
• aPRBind is applicable to modeled or unbound protein structures, expanding its utility.