Classification methods typically applied to the Invader assay include k-means clustering and the normal mixture model for original two-dimensional data or angle data. Combining the normal mixture model and angle data might result in an inproved method. In fact, such an approach has the advantages that it can be used to evaluate the goodness of classification for each individual and angle data are easily handled. However, the method requires that the data have an origin, which implies that one cluster must be specified before clustering. Therefore, an alternative method using the normal mixture model is desirable. We propose a mathematical model with a latent time variable. Optimization is based mainly on a one-dimensional normal mixture model with two components, which provides stable computational results more quickly than can be obtained using a bivariate normal mixture model.
2004 The Biometric Society of Japan