DPGOK: A Deep Learning-Based Method for Protein Function Prediction by Fusing GO Knowledge With Protein Features.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate protein function prediction is crucial for disease mechanism understanding and drug target discovery.
• Gene Ontology (GO) provides hierarchical and semantic context to improve prediction accuracy.
• Existing methods often fail to tailor GO embeddings to specific proteins, limiting functional relevance.

Purpose of the Study:

• To propose DPGOK, a deep learning method for protein function prediction using protein-aware GO embeddings.
• To enhance the integration of GO knowledge with protein sequence features for more accurate predictions.
• To generate protein-specific GO embeddings that reflect protein-functional relevance.

Main Methods:

• DPGOK employs deep learning to fuse protein features with protein-aware GO representations.
• GO semantic representations are learned using a knowledge graph loss.
• Protein-aware GO embeddings are generated guided by protein features.

Main Results:

• DPGOK significantly outperforms state-of-the-art methods across all Gene Ontology domains.
• The method effectively discovers hierarchically deeper and more informative protein functions.
• Knowledge graph loss improves the stability and semantic coherence of GO representations.
• Combining DPGOK with homology-based approaches further boosts predictive performance.

Conclusions:

• DPGOK offers a novel approach to protein function prediction by leveraging tailored GO embeddings.
• The method provides more accurate and informative functional predictions compared to existing techniques.
• DPGOK has the potential to advance disease mechanism studies and drug discovery efforts.