Identifying Protein-Nucleotide Binding Residues via Grouped Multi-task Learning and Pre-trained Protein Language...

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Accurate identification of protein-nucleotide binding residues is critical for understanding protein function and enabling drug discovery.
• Existing computational methods show promise but struggle with the diversity and limited availability of nucleotides.
• Predicting binding residues for a wide range of nucleotides remains a significant challenge in the field.

Purpose of the Study:

• To develop a novel computational approach, NucGMTL, for predicting protein-nucleotide binding residues across all observed nucleotides.
• To leverage deep multi-task learning and pre-trained protein language models to enhance prediction accuracy.
• To address the challenge of nucleotide variability by effectively utilizing shared binding patterns.

Main Methods:

• NucGMTL employs a grouped deep multi-task learning framework.
• It utilizes pre-trained protein language models for robust sequence embeddings.
• Multi-scale learning and scale-based self-attention mechanisms are incorporated to capture feature dependencies.

Main Results:

• NucGMTL achieved an average area under the Precision-Recall curve (AUPRC) of 0.594 on benchmark datasets.
• The method outperformed existing state-of-the-art approaches in predicting protein-nucleotide binding residues.
• The effectiveness stems from the integration of grouped multi-task learning and pre-trained protein language models.

Conclusions:

• NucGMTL represents a significant advancement in predicting protein-nucleotide binding residues for diverse nucleotides.
• The proposed grouped multi-task learning strategy effectively captures shared binding patterns.
• The method offers a valuable tool for protein function annotation and drug discovery efforts.