Annotating Proteins with Incomplete Label Information

• Published in: Pattern Recognition in Computational Molecular Biology pp. 585 - 608
• Main Authors: Yu, Guoxian, Rangwala, Huzefa, Domeniconi, Carlotta
• Format: Book Chapter
• Language: English
• Published: Hoboken, NJ, USA John Wiley & Sons, Inc 19.11.2015
• Subjects: multi‐label learning algorithms; partially annotated proteins; ProDM method; protein function prediction; protein‐protein interaction benchmarks
• ISBN: 9781118893685; 1118893689
• DOI: 10.1002/9781119078845.ch29

This chapter studies protein function prediction using partially annotated proteins. It reviews related work on multi‐label learning algorithms for network‐based protein function prediction and weak‐label learning approaches. The chapter then introduces protein function prediction using dependency maximization (ProDM) and details the experimental setup. It investigates the performance of ProDM on replenishing missing functions and predicting protein functions on three different protein‐protein interaction (PPI) benchmarks. The first data set, Saccharomyces cerevisiae PPIs (ScPPI), is extracted from BioGrid. The second data set, KroganPPI, is obtained from the study of Krogan et al. The third data set, HumanPPI is obtained from the study of Mostafavi and Morris. The experimental results demonstrate the benefit of integrating the guilt by association rule, function correlations, and dependency maximization in protein function prediction. In future work, it is planned to incorporate more background information of proteomic data and hierarchical structure among function labels for protein function prediction.