IEEE/ACM Trans Comput Biol Bioinform. 2018 Jul-Aug;15(4):1247-1258. doi: 10.1109/TCBB.2015.2476809. Epub 2015 Sep 25.
Proteins carry out their function in a cell through interactions with other proteins. A large scale protein-protein interaction (PPI) network of an organism provides static yet an essential structure of interactions, which is valuable clue for understanding the functions of proteins and pathways. PPIs are determined primarily by experimental methods; however, computational PPI prediction methods can supplement or verify PPIs identified by experiment. Here, we developed a novel scoring method for predicting PPIs from Gene Ontology (GO) annotations of proteins. Unlike existing methods that consider functional similarity as an indication of interaction between proteins, the new score, named the protein-protein Interaction Association Score (IAS), was computed from GO term associations of known interacting protein pairs in 49 organisms. IAS was evaluated on PPI data of six organisms and found to outperform existing GO term-based scoring methods. Moreover, consensus scoring methods that combine different scores further improved performance of PPI prediction.
蛋白质通过与其他蛋白质的相互作用在细胞中发挥其功能。生物体的大规模蛋白质-蛋白质相互作用（PPI）网络提供了静态但必不可少的相互作用结构，这是理解蛋白质和途径功能的有价值线索。PPIs 主要通过实验方法确定；然而，计算 PPI 预测方法可以补充或验证实验中确定的 PPIs。在这里，我们开发了一种从蛋白质的基因本体论（GO）注释中预测 PPIs 的新评分方法。与现有的将功能相似性视为蛋白质相互作用指示的方法不同，新的评分，称为蛋白质-蛋白质相互作用关联评分（IAS），是从 49 种生物体中已知相互作用蛋白对的 GO 术语关联计算得出的。IAS 在六种生物体的 PPI 数据上进行了评估，发现其表现优于现有的基于 GO 术语的评分方法。此外，组合不同评分的共识评分方法进一步提高了 PPI 预测的性能。
