Cluster Analysis Visualisations

A number of visualisation are provided for the examination of gene expression information

Cluster viewers are provided to examine differences/similarities between the clusters resulting from an individual analysis/algorithm.

Cluster comparison tool is provided to compare the results of two different cluster analyses or the results of a cluster analysis against a biological relevant categorization (e.g. cell cycle, disease state, regulatory networks information)

Cluster Comparison

This tool is designed for the comparison of two sets of results, for example comparing the results of a cluster analysis against some biological relevant categorized data set. The tool provides a graphical representation of the intersections between the different sets of clusters, and the (hyper geometrically defined) probability can be adjusted to show the rarer events (those which are most significant).

The example below shows the comparison of a Euclidian cluster analysis (with seeded values) against a prediction of yeast gene regulatory modules . The visualisation shows that for some regulatory modules the prediction using unsupervised techniques is surprisingly high (this is less likely to do with the predictive powers of the unsupervised technique and more likely to do with a similar geometric defined model being used to predict the gene regulatory modules).

Cluster Viewer

This tools is designed to give an indication as to the biological behaviour of the clusters found by SeqExpress. Some behaviours are relatively simple to ascertain by simple geometry (for example cyclic behaviours), however most genes have a complex involvement in the governing of cellular machinery and so further analysis is needed.

The figure above shows the basic layout of the tool, it consists of a number of features:

• Selected Menu: at the very top of the tools is a drop down menu for selecting the sets of clusters to be examined.
• Cluster List: at the top left of the tool is a list of clusters within the currently chosen set. One or more clusters can be selected from this list. The results of cluster validation studies (if available) are also shown.
• Group Cluster By Tools: on the top right of the tool are two buttons for the further analyses of clusters. These tools all perform hierarchical analysis of the clusters, but they each use a different distance function to calculate the pairs of items to cluster at each iteration.
  • Euclidian: Develops a hierarchy based on distance between the central points for each cluster. This will group the clusters that have simi liar values.
  • Cosine: Develops a hierarchy based on the angle between the average points of each cluster. This will group the clusters that respond by similar amount across the experiments.
• Scores are shown in the middle of the tool (if available).
• The parallel plot at the bottom of the tool shows the average profile for each cluster. This is useful in gaining some understanding of how the defined behaviour is mirrored by the actual expression values (the profile is the average distance, and so if the clustering method is not a distance or feature generated one, then such an average profile will not accurately represent the genes within the cluster).
  

Any selection within the cluster dendograms, parallel plots or lists within this tool are echoed in all the other views. Additionally selections in this tool are also echoed in any of the other open visualisations. If you wish to save the selected clusters as a new group of genes this can be done by selecting Edit->Copy Selected.