graphkernels

graphkernels computes graph kernels to quantify similarities between graph-structured biological data for use in machine learning and network analysis.

Key Features:

• Diverse Kernel Computations: Supports multiple graph kernels, including vertex and edge label histogram kernels; graphlet kernels, which capture local subgraph structures; random walk kernels as baseline methods; and the Weisfeiler-Lehman graph kernel.
• Efficiency: Core computations are implemented in C++ to provide high computational performance for large datasets typical in computational biology.
• Kernel Matrix Output: Computes kernel matrices that encode pairwise similarities between graphs for downstream machine learning tasks.

Scientific Applications:

• Protein-protein interaction network analysis: Quantifies similarities between protein-protein interaction (PPI) networks to support functional and comparative studies.
• Gene regulatory network comparison: Measures similarity between gene regulatory networks to aid analysis of regulatory relationships and network-level patterns.
• Machine learning on graphs: Supplies kernel matrices for use in classification, regression, and clustering of biological networks.

Methodology:

Transforms graph-structured data into kernel matrices using label-histogram, graphlet, random walk, and Weisfeiler-Lehman kernels, and uses those matrices in machine learning tasks such as classification, regression, and clustering.