Comparison of methods for identifying periodically varying genes.

Several methods have been reported for identifying periodically varying genes from gene expression datasets. We compare the performance of five existing methods and a combination of G-statistic and autocovariance (called GVAR) using simulated sine-function-based and cell-cycle-based datasets. Based on this analysis we recommend appropriate methods for different experimental situations (length of the time series, sampling interval and noise level). No single method performs the best under all tested conditions. None of the evaluated methods perform well at high noise levels for short time series data. At lower noise levels, GVAR performed the best.