NEW METHOD OF LATENT PERIODICITY DETECTION MAY DETERMINE STRUCTURALLY RELATED PROTEINS AND PROTEIN FAMILIES

Motivation: The study of repeats in biosequences is important because repeats in sequence often lead to repeats or regularities in structure. Recently we elaborated novel methods allowing locating subsequences in genetic texts that consist of hidden (non-homologous) repeats. But since the only criterion for the presence of latent periodicity (the property of a sequence composed of hidden repeats) is its statistical significance, we need to know whether other related sequences have this type of periodicity and thereby this periodicity pattern has some structural or functional sense. Results: We proposed and implemented the algorithm for cyclic alignment, which is a locally optimal alignment of a sequence against a cyclically elongated profile. Insertions and deletions are allowed in both sequence and profile, giving the algorithm more sensitivity to ancient and distant repeats where they are likely to arise. In addition, the profile can be trained to search for structurally and functionally related periodicity types. In our analysis of Swiss-Prot we found a number of previously unknown periodicity patterns corresponding to structural features of NAD-binding sites of dehydrogenases, active sites of protein kinases, dethiobiotin synthetases and others. For instance, a 24 residues long periodicity pattern turned out to correspond to Rossman-fold (or Rossman-like) domains. The analysis of periodicity profile showed that it contains alteration of α and β-structure preference, and this alteration is known to be common for nucleotide-binding domains. Availability: The software (C++ source and executables for different platforms) and detailed results for investigated types of periodicity are available upon request from the authors