Table 1.
Comparison of single rate versus random models for 3 alignments.
Figure 1.
Distribution of log likelihood scores for multi-rate models where amino-acid substitutions are assigned randomly to 5 non-synonymous classes.
The fit of single rate (SR), linear combination of amino acid properties (LCAP), empirical codon model (ECM), Genetic Algorithm (GA) and the general reversible model (REV) are shown as upside-down triangles. Number of rate classes inferred in the GA are 3, 4 and 5 for PF00803, rhodopsin and HIV-1 pol, respectively. Dashed lines indicate the log likelihood required to (i) reject the single rate model in favor of a 5 rate model (left), and (ii) reject a 5 rate model in favor of REV (right). All models were fitted using maximum likelihood estimates of position-specific nucleotide frequencies.
Table 2.
Comparison of empirical model fits using BIC.