SWIFT (sequence-wide investigation with Fourier transform): a software tool for identifying proteins of a given class from the unannotated genome sequence

Abstract

The ever increasing number of sequenced genomes calls for new analysis techniques, which can benefit from the methodologies developed in the field of signal processing. The present paper addresses the question of searching a pattern of amino acids (not necessarily completely specified) by means of the cross-correlation of complex sequences, obtained after suitable coding of the original amino acid sequence. Subsequently, the proposed algorithm provides a flexible strategy in setting the border between the accepted and rejected ORFs, by means of the k-means clustering of the candidate ORFs. The search for the class of proteins specified by the pattern is carried out from the most basic level, i.e. the DNA sequence, without sifting through an ensemble of previously determined ORFs. Thus, an exhaustive examination of all the occurrences of the pattern in the genome is performed. The application of the method to the search of surface proteins in Gram-positive bacteria witnesses its efficacy, in terms of both sensitivity and specificity. The comparison with the usual (and somewhat arbitrary) choice of setting a fixed value for the threshold length of the putative ORF confirms the validity of the proposed approach.