Figure 2

Statistical significance of Markov chain graph clustering results. The best p-values for over-representations of Gene Ontology (GO) terms, averaged over all clusters (S score, denoted by color arrow) are compared to the analogous values for 100 randomly-obtained clusterings (histogram). GO categories: (A) Molecular Function, (B) Biological Process, (C) Cellular Component. In each case, the actual clustering scored significantly better than any of 100 randomly obtained ones (Wilcoxon test p-value < 2.2 × 10^-16). In a comparison of the MCL (Markov clustering) and k-means clustering results (the latter denoted by black arrows), MCL had better S scores for GO terms overrepresentation than the k-means method (0.0016 versus 0.0020 for "Molecular Function" category; 0.0016 versus 0.0026 for "Biological Process"; and 0.0044 versus 0.0050 for "Cellular Component").