Structure-based prediction of nucleic acid binding residues by merging deep learning- and template-based approaches

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Accurate prediction of nucleic acid-binding residues is vital for understanding protein function in transcription and translation.
• Traditional hybrid methods are being surpassed by deep learning, but integrating deep learning with template-based approaches remains underexplored.

Purpose of the Study:

• To develop a novel structure-based integrative algorithm, NABind, for accurate prediction of DNA- and RNA-binding residues.
• To explore the integration of deep learning and template-based strategies for enhanced prediction performance.

Main Methods:

• Developed a deep learning module using sequence/structural descriptors and graph attention networks.
• Constructed a template module by transforming alignment features for supervised learning.
• Integrated deep learning and template modules using a stacking strategy, followed by random walk-based post-processing.

Main Results:

• NABind achieved excellent performance on both native and predicted protein structures.
• The algorithm outperformed existing hybrid and recent deep learning methods in predicting nucleic acid-binding residues.
• The NABind server is publicly available for use.

Conclusions:

• The novel NABind algorithm effectively integrates deep learning and template-based methods for superior nucleic acid-binding residue prediction.
• This integrative approach advances the accuracy of predicting key residues involved in transcription and translation.
• NABind offers a valuable tool for researchers studying protein-nucleic acid interactions.