Metabolic pathway membership inference using an ontology-based similarity approach

Abstract

Determining whether a protein belongs to a metabolic pathways is an important annotational task, can provide context to the basic functional annotation and aid reconstruction of incomplete pathways. In this work, we develop a method for pathway membership inference based gene ontology (GO) similarity between a query protein and proteins that are known to the members of a a given pathway. We specifically use human metabolic pathway from KEGG and human gene annotation dataset from Gene Ontology in this experiment. By comparing with various existing GO term semantic similarity, we develop an effective and efficient way to take into both information content of individual GO terms and the whole GO hierarchy. We test the classifier using 10-fold cross validation for all metabolic pathways reported in KEGG database and demonstrate that our method either outperform with statistical significance or perform comparably with a suite of existing semantic similarity measures, as evaluated using ROC score. And our method outperforms other methods in running time by multiple orders of magnitude for long pathways.