iMPTCE-Hnetwork: A Multilabel Classifier for Identifying Metabolic Pathway Types of Chemicals and Enzymes with a Heterogeneous Network

Figure 2

Entire procedures for constructing and evaluating the multilabel classifier to identify metabolic pathway types of chemicals and enzymes. The information of metabolic pathways is retrieved from KEGG, inducing the labels of chemicals and enzymes. Three types of interaction information are obtained from STITCH and STRING to construct a heterogeneous network. The network embedding algorithm, Mashup, is applied on the heterogeneous network to extract feature vectors of chemicals and enzymes. Labels and vectors are fed into the RAndom k-labELsets (RAKEL) algorithm, incorporating support vector machine (SVM) as the basic classifier, to construct the multilabel classifier. The classifier is evaluated by ten-fold cross-validation.