Statistical methods for detecting periodic fragments in DNA sequence data

摘要: Period 10 dinucleotides are structurally and functionally validated factors that influence the ability of DNA to form nucleosomes, histone core octamers. Robust identification periodic signals in sequences is therefore required understand nucleosome organisation genomes. While various techniques for identifying components genomic have been proposed or adopted, requirements such not considered detail confirmatory testing a priori specified periods has developed. We compared estimation accuracy suitability autocorrelation, discrete Fourier transform (DFT), integer period (IPDFT) previously Hybrid measure. A number different statistical significance procedures were evaluated but blockwise bootstrap proved superior. When applied synthetic data whose period-10 signal had eroded, which was approximately period-10, technique exhibited superior properties during exploratory estimation. In contrast, using procedure identified IPDFT as having greatest power. These on yeast defined from ChIP-chip study where metric confirmed expected dominance associated more significant occurrences period-10. Application whole genomes mouse ~ 21% 19% respectively these spanned by positioning (NPS). For estimating dominant period, we find method empirically be most effective both eroded approximate periodicity. The found measure, performing particularly well problem detection presence autocorrelation poorly suited use with bootstrap. our methods two model organisms revealed striking proportion NPS. Despite their markedly sizes, roughly equivalent proportions (19-21%) lie within spans NPS {AA, TT, TA}. biological regions remains demonstrated. To facilitate this, coordinates available Additional files 1, 2, 3 format suitable visualisation tracks popular genome browsers. This article reviewed Prof Tomas Radivoyevitch, Dr Vsevolod Makeev (nominated Mikhail Gelfand), Rob D Knight.