Graph kernels for chemoinformatics

The authors review graph kernels which is one of the state-of-the-art approaches using machine learning techniques for computational predictive modeling in chemoinformatics. The authors introduce a random walk graph kernel that defines a similarity between arbitrary two labeled graphs based on label sequences generated by random walks on the graphs. They introduce two applications of the graph kernels, the prediction of properties of chemical compounds and prediction of missing enzymes in metabolic networks. In the latter application, the authors propose to use the random walk graph kernel to compare arbitrary two chemical reactions, and apply it to plant secondary metabolism.