Molecular sequence accuracy and the analysis of protein coding regions.

Molecular sequences, like all experimental data, have finite error rates. The impact of errors on the information content of molecular sequence data is dependent on the analytic paradigm used to interpret the data. We studied the impact of nucleic acid sequence errors on the ability to align predicted amino acid sequences with the sequences of related proteins. We found that with a simultaneous translation and alignment algorithm, identification of sequence homologies is resilient to the introduction of random errors. Proteins with greater than 30% sequence identity can be reliably recognized even in the presence of 1% frameshifting (insertion or deletion) error rates and 5% base substitution rates. Incorporation of prior knowledge about the location and characteristics of errors improves tolerance to error of amino acid sequence alignments. Similarly, inclusion of prior knowledge of biased codon utilization by yeast (Saccharomyces cerevisiae) allows reliable detection of correct reading frames in yeast sequences even in the presence of 5% substitution and 1% frameshift errors.