Figure 5.

Codon context bias is organized in discrete groups. A two-way Pearson clustering by single linkage of the codon context data highlights regions of good and bad codon context, indicating that codon context bias is highly structured. A significant number of codons do not fall into the major clusters, indicating that their preferences and rejections are defined on a one-to-one basis. The 3' codon contexts whose residual values fall within the nonstatistically significant -3 to +3 interval are also scattered in the map, indicating that there is no cluster of codons that have little or no preference for particular codons as 3' neighbors.