Figure 1

Effect of sequence length on behaviour of codon usage measures. Figure 1a, 1c and 1e illustrate the degree of misestimation of a measure at varying sequence lengths (x axis), compared to the values at 2500 codons.. The values were obtained by calculating means for 10000 randomly generated sequences per method per length, and are expressed as percentages of the measures' dynamic range (see Methods). Figures 1b, 1d and 1f display standard deviations of the same 10000 measurements, as percentage of the dynamic range; higher values mean a method is more 'noisy'. In Figures 1a and 1b we generated sequences unbiased in use of codons and compared them to a frequency table also assuming equal use ('None', see Methods'). In 1c and 1d both the sequences and the expected frequency were, on overall, biased ('Med-1'); in 1e and 1f the sequences were biased ('Med-1'), but were compared to an unbiased expected frequency table ('None').