Ka Yee Yeung, Chris Fraley, Alejandro Murua, Adrian E. Raftery, Walter L. Ruzzo

• Pdf file of preprint (July 2001). (contain color figures) (Bioinformatics 2001, volume 17, number 10, pages 977-987) in conjunction with the 3rd Georgia Tech-Emory International Conference on Bioinformatics

• Technical Report UW-CSE-01-04-02 (April 2001) postscript pdf. These files contain color figures.

• Technical Report 396 in Department of Statistics (April 2001) postscript pdf. These files contain color figures.

• Most recent written description of this work (Dec 2001) in Chapter 5 of Ka Yee's thesis.

• Software: We used the MCLUST implementation to produce results for the EI (equal volume spherical), VI (variable volume spherical), VVV (unconstrained) and EEE models. The implementation for the diagonal model is not yet available.

• Note that in the Bioinformatics 2001 paper, the results were obtained by using the same model in both the hierarchical initialization and EM steps. For example, the results reported for the EI model were obtained by using EI in both the hierarchical and the EM steps.

• Supplement: Data transformations and the Gaussian mixture assumption postscript pdf. Since the model-based approach assumes that each of the clusters is generated by the multivariate Gaussian distribution, we tested the normality of each class in the data sets with external criteria.

• Enlarged and colored figures in July 2001 version

• Additional figures not shown in July 2001 version

• Gene Expression data sets used:
  • Unfortunately, the original web site containing the yeast cell cycle data by Cho et al. 1998 no longer seems to work. If you are interested in the full data, you can get the processed data from Spellman et al. or the raw data from SMD. In both cases, you would need to select the experimental conditions that you need.
    • Due to popular demand, we are making the subset of 384 genes we used in Ka Yee's dissertation available (as text-delimited file).
      • raw subset data without any normalization
      • log-transformed subset data
      • standardized subset data
      • legend.
    • We are also making the subset of 237 genes (corresponding to 4 MIPS categories) used in Ka Yee's dissertation available (as text-delimited file).
      • log-transformed
      • standardized
  • The ovary data set is not publicly available yet. Unfortunately, we do not have permission to distribute this data.

  ---

  Due to popular demand, this web site was updated on 3/24/2006.

  If you have any questions or comments on this paper, feel free to email Ka Yee

  Back to Ka Yee's research page.