## Protocol 55 Cluster analysis

Of the 454 clones, 441 clones showed a consistent up- or down-regulation at the three differentiation time points were included in a cluster analysis. We clustered the 441 datasets, each representing a single time course, by applying the k-means algorithm and using a refined Euclidean distance measure (Figure 5.4). This specifically takes into account the time dependence of gene expression changes. We performed k means clustering with k values ranging from 3 to 15 and found that for our dataset, k =10 is optimal to separate many clearly different dynamics without separating genes with too similar dynamics. The distance was defined as the weighted sum of k-means assignment and a similarity of shapes between cluster centers (gradient). The distance measure (D) was defined as follows: D(x,y) = a d1 (x,y)+(1-a) d2(x,y) where a is 0.5, x,y are the two gene profiles to be compared, d1 the Euclidean distance between x and y, and d2 the gradient of x and y. These calculations and the respective visualization were carried out using MATLAB (Version 6.0.0.88, Release12, MathWorks, MA).

Cluster 1

Cluster 2

Cluster 3

Cluster 4

Cluster 5

BDNF

Cluster 1

24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h

24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h

Cluster 6 Cluster 7 Cluster 8 Cluster 9 Cluster 10

BDNF

24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h

24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h 24 h 48 h 96 h

Cluster 1 234 5678910 Sum

BDNF 24 51 36 106 54 39 21 15 18 77 441

Intersection 21 35 12 55 33 22 3 1 6 61 249

Figure 5.4.

(A) Clusters of clones with similar expression changes over time. The dynamic behavior of the 441 clones that are consistently up- or down-regulated in both the BDNF and NT4 series is shown separately. For each series, 10 clusters were found. Clusters 1-5 contain up-regulated, and clusters 6-10 down-regulated clones. The x-axis depicts the three time points of differentiation. The y-axis shows relative fold changes, that is expression changes referred to the undifferentiated state (0 h). These relative numbers are estimated logarithmic fold changes. Note that y-axis values at 24 h which are different from the value 0 indicate that the expression of a clone had already changed between 0 h and 24 h. (B) Numbers of clones contained in each cluster of the NT4 and BDNF series. Copyright © 2004 by the Society for Neuroscience.

A microarray-based screening method for known and novel SNPs

Ena Wang and Francesco M Marineóla

## Post a comment