Looking at some more detailed codon usage tables, genes may be further clustered into three gene classes: Metabolic genes, highly expressed genes during exponential growth, and horizontal gene transfer. Looking at the original paper by Medique et al., they clustered the genes based on the CAI and then by a variant of k-means determined 3 classes. Note that this is different from a class II gene which is determined by the types of RNA polymerase used.
How did they end up determining what the three classes are? It seems as if they made this generalization without any proteome data. Would the same genes be classified using a protein expression data during exponential growth and stationary growth?