Incorporating chirality within the graph kernel framework

Molecules being often described using a graph representation, graph kernels provide an interesting framework which allows to combine machine learning and graph theory in order to predict molecule’s properties. However, some of these properties are induced both by the covalent bound relationships between atoms and by constraints on the relative positioning of these atoms. Graph kernels based solely on the graph representation of a molecule do not encode the relative positioning of atoms and are consequently unable to predict accurately molecule’s properties connected with this relative positioning. In this report, ordered structured object are introduced in order to incorporate spatial constraints within the graph kernel framework. The incorporation of this new features within the graph kernel framework allows to predict accurately chiral information hence overcoming the previous limitation.