Cancer Diagnostics with DNA Microarrays (Knudsen/Cancer Diagnostics with DNA Microarrays) || Cluster Analysis

✍ Scribed by Knudsen, Steen

• Book ID: 101316305
• Publisher: John Wiley & Sons, Inc.
• Year: 2005
• Tongue: English
• Weight: 194 KB
• Edition: 1
• Category: Article
• ISBN: 0471784079
• DOI: 10.1002/0470041102.ch5

✦ Synopsis

If you have just one experiment and a control, your first data analysis will limit itself to a list of regulated genes ranked by the magnitude of up-and downregulation, or ranked by the significance of regulation determined in a t-test.

Once you have more experiments-measuring the same genes under different conditions, in different mutants, in different patients, or at different time points during an experiment-it makes sense to group the significantly changed genes into clusters that behave similarly over the different conditions. It is also possible to use clustering to group patients into those who have a similar disease. Clustering is often used to discover new subtypes of cancer in this way.

5.1 HIERARCHICAL CLUSTERING

Think of each gene as a vector of N numbers, where N is the number of experiments or patients. Then you can plot each gene as a point in N -dimensional hyperspace. You can then calculate the distance between two genes as the Euclidean distance between their respective data points (as the square root of the sum of the squared distances in each dimension).

This can be visualized using a modified version of the small example dataset applied in previous chapters (Table 5.1). The measured expression level of the five genes can be plotted in just two of the patients using a standard x -y coordinate system (Figure 5.1 left).

You can calculate the distance between all genes (producing a distance matrix ), and then it makes sense to group those genes together that are closest to each other in space. The two genes that are closest to each other, b and d, form the first cluster (Figure 5.1 left). Genes a and c are separated by a larger distance, and they form a Cancer Diagnostics with DNA Microarrays,