Summary: We propose a three-step periodicity detection algorithm named LSPR. Our method first preprocesses the raw time-series by removing the linear trend and filtering noise. In the second step, LSPR employs a Lomb-Scargle periodogram to estimate the periodicity in the time-series. Finally, harmonic regression is applied to model the cyclic components. Inferred periodic transcripts are selected by a false discovery rate procedure. We have applied LSPR to unevenly sampled synthetic data and two Arabidopsis diurnal expression datasets, and compared its performance with the existing well-established algorithms. Results show that LSPR is capable of identifying periodic transcripts more accurately than existing algorithms.
Bibliographical noteFunding Information:
Funding: Ministry of Science and Technology of China (2006CB100105); College Student Research and Career-creation Program of Beijing (2010).