A summary of three case studies based on microarray data analysis is presented in this poster: the application of supervised clustering techniques to gene-expression data, the application of unsupervised clustering techniques to gene-expression data, and methods for comparing different clustering algorithms.

The yeast expression database we used are the results of De Risi, et.al. Science, 278, 680-686 (1997). In this study, two yeast cell populations were studied, one growing anaerobically (reference), the other having undergone a diauxic shift, from an anaerobic to an aerobic metabolism.
The object of the study was to assay for changes in gene expression over the time course of the diauxic shift and to correlate the gene subsets whose synthesis was induced or repressed with the known enzyme metabolic pathways involved in the diauxic shift.

Distinguishing Exon from Intron DNA sequences is one of the goals of the Human Genome Project. We have encoded and classified human sequences from the Fickett dataset analytically and visually.
Several visualization and data mining techniques were used to validate and attempt to discover new methods for distinguishing coding DNA sequences, or exons, from non-coding DNA sequences, or introns.

John Weinstein's group at NCI has carried out an analysis of the GI50 (concentration at which 50% growth inhibition is observed) values of 141 chemical agents of known Mechanism Of Action (MOA) screened individually on 60 cells lines. GI50 values were used to classify the 141 chemicals into their appropriate MOA by applying a neural net classifier, achieving a 91.5% accuracy. We demonstrate similar neural net classificaiton accuracy using the GI50 data, as well as clustering of 6 MOA groups using GI50 and chemical property descriptor databases.

We developed a software platform in which large DNA sequence datasets may be visualized by techniques which readily reveal patterns and insights. Initially we focused on providing accurate statistical visualizations rather than quantitative presentations. An example application of this platform visualizes properties of DNA sequence strings of any size as a function of string position (for example, in large chromosome).

We have developed an experimental visualization environment (Exvis) that extends the current methods of visualization considerably and allows a scientist user to look at data and databases in a graphical manner whether or not the database is graphical in nature.
This technique extends the concept of mapping data values from a pixel-based approach to one using erceptually-based displayable objects, including auditory representation of data.
This is a much different approach than automated data mining, where algorithms determine relationships and structure in data. Exvis was meant to be used in an exploratory data analysis mode, where data can be rapidly and easily manipulated, organized into perceptually compelling displays, with flexibility for changing viewing parameters.

We have developed a multiple color model image display system that provides the capability to encode multiparameter images using several color models. This system allows interactive comparison and exloration of the potential merits of different color models for coding multiparameter image data into integrated displays. We also developed the color icon, which subsumes standard color coding techniques and retains their power, but escapes their limitation to three parameters. The color icon uses three human perception channels: color, texture and shape. Initial evaluations have demonstrated the power of this approach, which currently has the capability to encode up to 24 parameters.