Abstract

How to correctly and efficiently map small molecule to its possible metabolic pathway is a meaningful topic to metabonomics research. In this work, a novel approach to address this problem was introduced to encode physicochemical properties of small molecules. Based on this encoding method, a two stage feature selection method called mRMR-FFSAdaBoost was adopted to map small molecules to their corresponding metabolic pathways possible. As a result, the accuracies of 10-folds cross-validation test and independent set test for predicting the metabolic pathways of small molecules reached 83.88% and 85.23%, respectively. An online server for predicting metabolic pathways of unknown small molecules as described in this paper is accessible at http://chemdata.shu.edu.cn:8080/PathwayPrediction/.

Keywords: ChemAxon, jchem for excel, KEGG, metabolic pathway, online server, small molecules, web-server, biochemical reactions, living systems, biological functions.