AptRank: an adaptive PageRank model for protein function prediction on bi-relational graphs

• DOI: 10.1093/bioinformatics/btx029
• Publication Date: 2017-02-15T08:58:08Z
• Authors: Biaobin Jiang; Kyle Kloster; David F Gleich; Michael Gribskov
• Keywords: Protein function prediction; PageRank; Predictability

Diffusion-based network models are widely used for protein function prediction using data and have been shown to outperform neighborhood-based module-based methods. Recent studies that integrating the hierarchical structure of Gene Ontology (GO) dramatically improves accuracy. However, previous methods usually either GO hierarchy refine results multiple classifiers, or flattened into a function-function similarity kernel. No study has taken account together with as two-layer model.We first construct Bi-relational graph (Birg) model comprised both protein-protein association networks. We then propose two diffusion-based methods, BirgRank AptRank, which use PageRank diffuse information on this model. is direct application traditional fixed decay parameters. In contrast, AptRank utilizes an adaptive diffusion mechanism improve performance BirgRank. evaluate ability predict yeast, fly human datasets, compare four methods: GeneMANIA, TMC, ProteinRank clusDCA. design different validation strategies: missing prediction, de novo guided newly discovered comprehensively predictability all six find especially in when only 10% training.The MATLAB code available at https://github.rcac.purdue.edu/mgribsko/aptrank [e-mail] Bioinformatics online.