Limitations of codon adaptation index and other coding DNA-based features for prediction of protein expression in Saccharomyces cerevisiae.

The relationship between codon usage and protein/mRNA expression in S. cerevisiae has been extensively studied. Recently, protein expression data for the whole yeast genome was published. We investigate which properties of coding DNA sequences can be used to predict expression levels. The new algorithm by Carbone et al. for computing dominating codon bias in a genome is evaluated. It is concluded that it works at least as well as existing methods, and eliminates the need to arbitrarily choose a set of highly expressed genes. Also, the hypothesis that information on codon pair frequencies can be used to predict expression is investigated. Our conclusion is that codon pairs do not contribute more information than do single codon frequencies. Overall correlation between predicted and actual expression data using properties of coding DNA sequences is around 0.65. Hence, while being a useful source of information, the expression levels predicted by these methods should only be used as a rule of thumb. Copyright (c) 2004 John Wiley & Sons, Ltd.