DA-HGL: a domain-augmented heterogeneous graph learning framework for protein function prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate protein function prediction is vital for understanding diseases and advancing precision medicine.
• Current methods face challenges with sparse annotations and integrating multimodal data.
• Existing approaches often fail to holistically integrate diverse biological data sources.

Purpose of the Study:

• To develop DA-HGL, a novel heterogeneous graph learning framework for enhanced protein function prediction.
• To address the limitations of existing methods in handling annotation sparsity and multimodal data integration.
• To improve the accuracy and robustness of protein function prediction, especially for proteins with limited annotations.

Main Methods:

• Utilizing a heterogeneous graph learning framework (DA-HGL) that integrates protein sequences, domain architectures, and Gene Ontology (GO) hierarchies.
• Employing a multilayered graph structure and non-negative matrix factorization with dual biological constraints.
• Modeling domain-function coherence, GO semantic consistency, and topological congruence within the graph framework.

Main Results:

• DA-HGL demonstrated significant performance gains over state-of-the-art methods, with Fmax increases of 9.0% (yeast CC) and 17.2% (human BP).
• The framework successfully addresses annotation sparsity, showing particular strength in cold-start scenarios.
• DA-HGL accurately predicts functions for specific disease-related pathways, such as Parkinson's "ubiquitin-dependent catabolism".

Conclusions:

• DA-HGL provides a robust and effective framework for accelerating functional genomics and precision medicine.
• The method's ability to integrate multimodal data and handle sparse annotations offers a significant advancement in protein function prediction.
• This approach holds promise for deciphering complex disease mechanisms and enabling more targeted therapeutic strategies.