Chemical molecules or bio-molecules can be delineated in the form of a graph. Some of them can be active/inactive, toxic/non-toxic and resistant to a drug/not resistant to a drug. These properties of the molecules can be due to presence of sub structure inside the molecule which causes it to be toxic or non-toxic. So with this help of interpretability of the graph kernel, it will be easy to identify the substructure that has put this graph in a particular class. Moreover when a new bio-molecule or chemical molecule is introduced they can be easily classified as toxic/non-toxic on the basis of presence of that molecule. Thus instead of merely classifying the graph, if the substructures which lead to classification are available and interpretable, then it would be better. The presence of certain substructures that cause toxicity in a chemical molecule, has to be checked. To do this, the sparsity is impor­tant because when more and more substructures are taken for classification, it becomes computational expensive. Thus it is necessary to cherry pick the best molecule for the classification of the graph. The main objective is to classify the graph by extracting the most important substructure. In case of multiple important substructures, we use weighted graph classification. The technique is a quick feature extraction method based on the graph isomorphism Weisfeiler-Lehman isomorphism test. The algorithm involves running h iterations of WL-Subtree which in turn means a length feature of color hash for each node. This feature map will basically indicate the similarity of nodes at various depths for two different graphs. We also use the graph edge or node deletion that can be seen as a kind of edit operation.