Table 2 Summary of the simulation results when detecting the true number of subpopulations K (even sample sizes)
MethodaScenario 1Scenario 2Scenario 3
CONE (Mod)10 96% (0.20) 10 1110 98% (0.14) 10 1110 94% (0.24) 10 11
CONE (StARS)10 100% (0.00) 10 1010 100% (0.00) 10 1010 92% (0.37) 8 10
sNMF (CE)10 68% (0.79) 10 1210 84% (0.63) 10 1210 84% (0.66) 10 13
CONE (subsample)10 92% (0.51) 10 1210 96% (0.20) 10 1110 94% (0.24) 9 10
  • Island model (scenario 1), hierarchical island model (scenario 2), and hierarchical stepping-stone model (scenario 3). Mod, modularity; CE, cross-entropy.

  • a We have reported the median, percentage of the number of correctly inferred subpopulations, an error measure (RMSE), and minimum and maximum (in boldface) values of the number of inferred subpopulations. We used both the modularity approach and StARS to choose a proper value for the tuning parameter λ. We have also reported the number of inferred clusters identified with the StARS subsampling (subsample).