Extended Affinity Propagation

This step is optional.

This step determines similarities between shapes. In this step, similarities between shapes are determined. For pre-processing run 5-Extended_Affinity_Propagation > Run_Affinity_Propagation.m

Note: this step first needs to load CellShapeData.mat. When this file is large, loading the file may take a long time. Once loaded, the program prints progress on the console.

After successful completion, the following nine files should have been generated in the Analysis folder:

• avail_list.mat
• APclusterOutput.mat
• double_sim_matrix.mat
• linkagemat.mat
• pref.mat
• resp_mat.mat
• single_sim_matrix.mat
• unconv.mat
• wish_list.mat

Inspecting Automatic Classification of Shapes

To inspect the automatic classification, run 5-Extended_Affinity_Propagation > Inspect_ShapeAffinity.m, select the Analysis folder, and then the program produces two figures (see below)

The left figure shows the unordered exemplars with one example highlighted for each of 10 clusters, while the right figure shows ten example shapes from each cluster.

Dendrogram of Shapes

Often, the program produces to many clusters and it is desirable to constraint the number of clusters. To make an informed choice, the cluster dendrogram should be consulted. Run 5-Extended_Affinity_Propagation > Run_Dendogram_of_Shapes.m, select Analysis folder, and then the program produces the cluster dendrogram (see below). Use this to decide how many clusters are appropriate for the dataset.

Constrained Clustering

Then run 5-Extended_Affinity_Propagation > Run_constrained_Clustering.m, select the Analysis folder and configure the number of clusters based on your choice made above (see figure below).

Note: The number you enter for constraint clusters must be smaller than the number of exemplars from the affinity propagation; otherwise the program terminates without results.

Left figure shows all shapes in shape space coloured in the cluster colours. Each dot in the figure corresponds to a cell shape. The x and y axes are the first two Diffusion Coordinates. The right figure the ordered list of exemplars coloured in the cluster colours.