Fig. 2

GC3 and AU3 codons are not distributed randomly across coding sequences. a. The proportion of the GC3 codons of alanine, calculated as proportion of all codons for alanine in each coding sequence from the human genome plotted against the proportions of either the GC3 or the AU3 codons of proline (blue) or glutamate (purple). Each point represents a different coding sequence (different gene). This implies that if a sequence contains high percentages of GC3 codons for alanine, it will also tend to have high percentages of GC3 codons for either proline or glutamate. b. Correlation coefficients resuming all the possible combinations of codons (all amino acids against each other). All GC3 codons within a transcript positively correlate with all other GC3 codons. The same is true for AU3 codons. On the contrary, all GC3 codons anti-correlate with all AU3 codons. c. The same analysis as in b, performed after splitting the coding sequences in halves, and calculating the correlations between codons from one half, and codons from the other half. The observation from B remains valid even in these conditions