Cells change their own states, stopping replication or repressing protein syntheses in order to adapt to the external environment such as temperature change, lack of nutrition or other stressful stimuli. As a result of these changes, they transform from the exponential growth phase into a stationary phase or into an anaerobic condition, etc. In conventional studies of microarray time series analyses, a fixed transcriptional regulatory formation is assumed through all data series. On the contrary, we assume that the regulatory formations can be changed, and estimate the transition time points from a statistical viewpoint. Using a linear dynamical system and microarray data, we estimate cellular internal states that are unobservable by experiments and detect transitions of internal states based on the temporary descent of log-likelihood values. Combining the results with the classification results based on a self-organizing map, significant cellular transition points can be detected and activation of particular functional genes concerned with each state of transition extracted. This approach gives us an objective standard for understanding cellular changes and the estimation of unknown gene functions in microarray time series analyses.
