Accounting for background nucleotide composition when measuring codon usage bias.