Codon Adaptation Index Calculator

The Codon Adaptation Index scores a coding sequence against the codon preferences of a chosen host, on a scale from just above 0 to 1. Its input is a table of relative adaptiveness values, written w, built from a reference set of very highly expressed genes in that organism: for each amino acid the most frequently used synonymous codon is assigned w of 1, and every alternative gets the ratio of its own frequency to that maximum. CAI is then the geometric mean of the w values along the gene, computed as the exponential of the mean natural logarithm. The geometric mean is deliberate, since a single codon with very low w drags the score down far more than an arithmetic average would, which is the intended behavior when rare codons cause ribosome pausing. Methionine and tryptophan, having one codon each, carry no information, and stop codons are excluded. Interpretation is organism-relative but broadly consistent. In Escherichia coli, ribosomal proteins and glycolytic enzymes typically score above 0.7, while poorly expressed regulatory genes fall near 0.2 to 0.3; for heterologous expression a redesigned gene is usually targeted above 0.8 in the host. The score is only as good as its reference table, and tables from the Kazusa database differ between strains and between expression conditions. More importantly, CAI predicts expression imperfectly because it ignores everything except codon identity: mRNA secondary structure near the start codon, the strength of the Shine-Dalgarno or Kozak context, transcript stability, and codon pair bias all matter independently. There is also evidence that some rare codons are functional, slowing translation at domain boundaries to aid folding, so driving CAI to its maximum can lower soluble yield even as it raises the index.

The idea that synonymous codons are used unevenly emerged as soon as gene sequences became available in the late 1970s. Richard Grantham and colleagues at Lyon proposed the genome hypothesis in 1980, arguing that each species has a characteristic codon usage signature. Toshimichi Ikemura, working in Japan, supplied the mechanism in a series of papers beginning in 1981, showing that in Escherichia coli and later in yeast the preferred codons correspond to the most abundant tRNA species, and that the correspondence is strongest in the most highly expressed genes. CAI itself was defined by Paul Sharp and Wen-Hsiung Li in Nucleic Acids Research in 1987, in a paper that introduced both the relative adaptiveness weighting and the geometric mean formulation, and demonstrated that the index correlates with measured protein abundance in E. coli. Its adoption was rapid because it condensed a whole codon table into one interpretable number. Related measures followed and are often reported alongside it, notably the effective number of codons devised by Frank Wright in 1990, which quantifies bias without needing a reference gene set. Practical importance grew sharply in the 1990s and 2000s as recombinant protein production moved routinely across species boundaries and commercial gene synthesis made whole-gene redesign affordable; codon optimization services now report CAI as a standard deliverable. Reference tables were centralized in the Codon Usage Database maintained at Kazusa DNA Research Institute, compiled from GenBank by Yasukazu Nakamura and colleagues, which remains the usual source of the w values such calculations depend on.