Codon Adaptation Index (CAI) Codon adaptation index is a measurement of the relative adaptiveness of the codon usage of a gene towards the codon usage of highly expressed genes. The relative adaptiveness (w) of each codon is the ratio of the usage of each codon, to that of the most abundant codon for the same amino acid. The CAI index is defined as the geometric mean of these relative adaptiveness values. Non-synonymous codons and termination codons (dependent on genetic code) are excluded. CAI values range from 0 to 1, with higher values indicating a higher proportion of the most abundant codons. [Sharp, P. M., and W. H. Li , (1987). The codon adaptation index a measure of directional synonymous codon usage bias, and its potential applications. Nucleic Acids Research 15: 1281-1295. Abstract, also see: Jansen R., Bussemaker H.J., and Gerstein M. (2003) Revisiting the codon adaptation index from a whole-genome perspective: analyzing the relationship between gene expression and codon occurrence in yeast using a variety of models. Nucleic Acids Res. 31(8):2242-51. Abstract ]

You've answered your question in your post. CAI measures the percentage of codons that are the most abundant choice in any organism. For instance, here are the choices for Isoleucine in E coli from the Codon Usage Database:

AUU Ile 0.50
AUC Ile 0.40
AUA Ile 0.09

A value of 1.0 would indicate all Ile codons are the most abundant (AUU). As you move away from 1.0 you have increasingly more of the AUC and AUA choices.

If you are seeing higher CAI values in your test set compared to the reference, this means they have a higher proportion of the most used codon choices.

CAI is often equated to expression (high CAI = high expression) although the relationship is not that simple. This graphic from DNA 2.0 (a gene synthesis company) demonstrates cases where CAI does not correlate with gene expression.

To understand more about CAI check out Frank Wright's earlier work on the Effective number of codons

Codon Adaptation Index (CAI) was first introduced by Sharp and Li to measure synonymous codon usage bias for a DNA or RNA sequence.It also measures the resemblance between the synonymous codon usage of a gene and the synonymous codon frequencies of a reference set.CAI was originally proposed to provide an estimate that can be used across genes and species, ranging from 0 to 1.

If a gene always uses the most frequently used synonymous codon in the reference set,then CAI=1. If a gene always uses the least frequently used synonymous codon in the reference set,then CAI=0.

Click on this link to explore more:Implementation Of Codon Adaptation Index (CAI) Using Biopython