AUTHOR=Zhou Jiaogen , Xiong Wei , Wang Yang , Guan Jihong 

TITLE=Protein Function Prediction Based on PPI Networks: Network Reconstruction vs Edge Enrichment

JOURNAL=Frontiers in Genetics

VOLUME=Volume 12 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/genetics/articles/10.3389/fgene.2021.758131

DOI=10.3389/fgene.2021.758131

ISSN=1664-8021

ABSTRACT=Over the past decades, massive amounts of protein-protein interaction (PPI) data have been accumulated due to the advancement of high-throughput technologies. Accordingly, a number of approaches based on PPI networks have been proposed for protein function prediction. Unfortunately, these PPI networks face serious data quality challenges, including high false positives (noise) and high false negatives (incompleteness), which adversely affect the performance of protein function prediction.
    To boost the performance of protein function prediction based on PPI networks, two major types of approaches were proposed to construct more robust and reliable PPI networks, including network reconstruction and edge enrichment. Although various implementations of the two types of approaches have been reported, and definite performance improvements have been achieved, there still lacks a systematic performance comparison between these two types of approaches. To answer this question, this study carried out a comprehensive performance comparison of these two types of approaches. Concretely, we first reconstructed and enriched PPI networks by using protein sequence similarity, local similarity indices and global similarity indices, and then compared the prediction performance of these reconstructed and enriched networks with that of the original networks of two real PPI datasets.
   From the experimental results, the enriched networks achieved more accurate predictions than the original networks and the reconstructed networks. The reconstructed network performed better than the original network of the BioGRID dataset that contains only physical interactions. However, the original network of the STRING dataset that contains known and predicted interactions outperforms the reconstructed network. Moreover, the sequence similarity was more effective than global similarity and local similarity in PPI network enrichment. In summary, we concluded that edge enrichment performed better than network reconstruction on current PPI networks, while network reconstruction was more effective on relatively small and incomplete PPI networks. As the PPI networks became more and more complete, the effectiveness of both edge enrichment and network reconstruction would decrease or relatively decrease.