Locally sensitive backtranslation based on multiple sequence alignment

Backtranslation is the process of decoding an amino acid sequence into a corresponding nucleic acid. Classical methods are based on the construction of a codon usage table by clustering and detection of the most probable codon used for each amino acid. We present a new method for backtranslation which is sensitive to the local position of the amino acid in the input sequence. The method makes use of multiple sequence alignment of the set of proteins under analysis. A local codon usage table stores for each amino acid X and for each position of X in the alignment the most used codon. We compared our method with EMBOSS using both ClustalW and AntiClustAl for multiple sequence alignment. Experiments showed that our method outperforms EMBOSS in terms of precision of backtranslation: the matching between the proteins obtained by our method and the original protein templates is clearly superior to that obtained by EMBOSS. This enforces the validity of a locally sensitive approach.