Gene set sample size correlates with codon usage variability.

<p>(<b>A</b>) Genes were randomly sampled from the human genome to create sets of different sizes corresponding to GO term set sizes (x-axis). The codon usage of each set was calculated and correlated with the exonic background (Spearman’s rank correlation coefficients (ρ); y-axis). For each set size, the distribution of 10 000 repeated samples is summarized in the figure by a vertical segment connecting the 1^st with the 3^rd quartile (interquartile range, IQR); the median is indicated as a black dot; the 2^nd and 98^th percentile are indicated with smaller dots. (<b>B</b>) For each human GO gene set, the Spearman correlation between its codon usage and the codon usage of the exonic background is plotted (y-axis) against the number of genes in this set (x-axis). Each point in the graph corresponds to a GO term. For all gene set sizes, the distributions from (A) were used as a background to empirically test the null hypothesis that the codon usage correlation of a GO term is explained by random sampling alone. GO terms whose codon usage significantly diverges from the background (FDR-adjusted <i>p</i> < 0.05) are colored in red; GO terms without significant divergence are colored in black. The lines indicate the 2^nd and 98^th percentile of the random distribution from (A).