The Graphlet Kernel (Shervashidze et al., 2009) measures graph similarity by comparing the frequency distributions of small connected non-isomorphic subgraphs called graphlets. There are 29 graphlets on 2–5 nodes, yielding a 73-dimensional signature vector per graph. Two graphs are considered similar if their normalized graphlet degree distributions (GDD) are close — the kernel value is their inner product. Unlike Weisfeiler-Lehman or random walk kernels, graphlet kernels capture local topology without label information, making them powerful for unlabeled biological networks where subgraph patterns encode function.