Fuzzy clustering of time series gene expression data with cubic-spline

Data clustering techniques have been applied to extract information from gene expression data for two decades. A large volume of novel clustering algorithms have been developed and achieved great success. However, due to the diverse structures and intensive noise, there is no reliable clustering approach can be applied to all gene expression data. In this paper, we aim to the feature of high noise and propose a cubic smoothing spline fitted for the time course expression profile, by which noise can be filtered and then groups genes into clusters by applying fuzzy cmeans clustering on the resulting splines (FCMS). The discrete values of radius of curvature are used to compute the similarity between spline curves. Results on gene expression data show that the FCMS has better performance than the original fuzzy c-means on reliability and noise robustness.