ΔBIC reports the improvement by the best model found by the genetic algorithm search compared to the baseline MG94 model. The 2 × 2 contingency table of simulated/inferred classifications for each codon pair, the corresponding P-value for association, and the MCC are provided. Bolded values in the 2 × 2 contingency table highlight the number of simulated codon classes that were correctly classified by the genetic algorithm. For classification purposes, we place each codon pair in the selected/neutral class based on which received more model averaged support. To quantify the degree of assignment uncertainty, we also tabulate how many of the 67 rates were “ambiguous,” i.e. had <0.9 support for the class assignment.