Protein Function Prediction Using GO Similarity-based Heterogeneous Network Propagation

Background: Protein function prediction serves as a fundamental cornerstone in bioinformatics, offering critical insights into the intricate biological processes and molecular mechanisms that form the basis of life. Precise annotation of protein functions is indispensable for unraveling disease mechanisms, identifying drug targets, and propelling forward synthetic biology applications. Nevertheless, this task remains complex due to the diverse characteristics of multi-omics data and the hierarchical structure of Gene Ontology (GO) annotations. Results: To tackle these challenges, we have developed an innovative approach that seamlessly integrates the topological structure of protein-protein interaction networks, a wide array of biological data including protein domain profile and protein complex information, and Gene Ontology into a heterogeneous network based on GO similarity. This integrated network encapsulates the multifaceted relationships between proteins and their functional annotations, setting the stage for a comprehensive protein function prediction framework. Building on this heterogeneous network, we have devised a protein function prediction method named GOHPro, which leverages the strength of network propagation algorithms. To evaluate the effectiveness of GOHPro, we conducted rigorous experiments on two model organisms, yeast and human, using the most up-to-date GO annotations dataset and the CAFA3 dataset. Our method was compared against several state-of-the-art approaches, and the results unequivocally showed that GOHPro outperforms its competitors, highlighting its superiority in predicting protein functions. The code and dataset of GOHPro are freely available at https://github.com/husaiccsu/GOHPro. Conclusions: The proposed GOHPro method, which seamlessly combines multi-omics data, protein interaction networks, and GO annotations within a heterogeneous network framework, significantly enhances protein function prediction accuracy. The outstanding performance of GOHPro in our experiments underscores its potential as a powerful tool for annotating protein functions, facilitating a more profound understanding of biological processes and contributing to advancements in bioinformatics and computational biology.