CLUSPECT is a new clustering method applicable for microarray data (Tritchler et al, 2005). This paper introduces a clustering method for microarray data based on eigenanalysis. The method is computationally efficient and can be interpreted in terms of familiar statistical models. Using simulation studies, we demonstrated that our spectral clustering method outperformed the two most widely used clustering methods, namely hierarchical and k-means (Tritchler et al, 2005). For example, comparing truth with the outputs of k-means, spectral and hierarchical clustering for a moderately difficult clustering problem, we obtained average adjusted Rand indices of 0.77 for k-means, 0.74 for hierarchical and 0.98 for spectral. The adjusted Rand statistic that was used to compare the performance of the three clustering methods can range from 0 to 1, with 1 being perfect agreement. Our method can incorporate supervision, in order to produce clusters whose variation can predict clinical outcome.
Sebastian Hirjoghe and David Tritchler
:: MORE INFORMATION
Tritchler, D., Fallah, S., and Beyene, J. (2005).
A spectral clustering method for microarray data.
Computational Statistics and Data Analysis, 49:63-76.