Clustering short time series gene expression data

Motivation: Time series expression experiments are used to study a wide range of biological systems. More than 80% of all time series expression datasets are short (8 time points or fewer). These datasets present unique challenges. On account of the large number of genes profiled (often tens of thousands) and the small number of time points many patterns are expected to arise at random. Most clustering algorithms are unable to distinguish between real and random patterns. Results: We present an algorithm specifically designed for clustering short time series expression data. Our algorithm works by assigning genes to a predefined set of model profiles that capture the potential distinct patterns that can be expected from the experiment. We discuss how to obtain such a set of profiles and how to determine the significance of each of these profiles. Significant profiles are retained for further analysis and can be combined to form clusters. We tested our method on both simulated and real biological data. Using immune response data we show that our algorithm can correctly detect the temporal profile of relevant functional categories. Using Gene Ontology analysis we show that our algorithm outperforms both general clustering algorithms and algorithms designed specifically for clustering time series gene expression data.