Detecting Periodically Expressed Genes based on Time-frequency Analysis and L-curve Method

In microarray experiments, gene expression profiles are often affected by biological properties, such as synchronization loss, and show some non-stationarity. Worse still, the microarray data usually suffers from missing values. The conventional spectrum-based methods, when used to identify a subset of genes that are periodically expressed, are degraded by these factors. In this paper, we use the Wigner-Ville distribution analysis and L-curve method for detection of periodically expressed genes. We provide a graphical exploratory device for assessment of the presence of periodically expressed genes. Then, we identify the subset of genes actually involved in the cell cycle using the L-curve method. The experiments on several widely used datasets show that our algorithm can effectively reduce the effect of non-stationarity and missing values problems.