Performance Analysis of Graph Laplacian Matrices in Detecting Protein Complexes
Detecting protein complexes is an important way to discover the relationship between network topological structure and its functional features in Protein-Protein Interaction (PPI) network. The spectral clustering method is a popular approach. However, how to select its optimal Laplacian matrix is still an open problem. Here, the authors analyzed the performances of three graph Laplacian matrices (unnormalized symmetric graph Laplacians,, normalized symmetric graph Laplacians and normalized random walk graph Laplacians, respectively) in yeast PPI network. The comparison shows that the performances of un normalized and normalized symmetric graph Laplacian matrices are similar, and they are better than that of normalized random walk graph Laplacian matrix. It is helpful to choose proper graph Laplacian matrix for PPI networks' analysis.