Figure 5. Comparison of different clustering methods: k-means clustering versus hierarchical clustering, combined versus separate anatomic and electrophysiological analysis. A, UPGWA hierarchical clustering using combined anatomic and electrophysiological data yielded similar results to k-means clustering. Ai, UPGWA hierarchical dendrogram separates the 106 interneurons sequentially by the least squared Euclidean distance. Each branching point represents the splitting of a cluster into two clusters, until the clusters are comprised of single neurons. Each end point thus represents a single interneuron. Branch points above the height of 2.8 (a.u.), in this case representing 52% of the maximum distance in the population, are considered to represent distinct clusters. These resulted in eight different clusters. To match up these clusters with those derived from the k-means clustering analysis, all possible permutations were tested. The permutation with maximum overlap, shown per the labels for each cluster, was used for further analysis. Aii, The clustering distribution for hierarchical clustering is shown on the left bar, with each color corresponding to the branch on the dendrogram. The four clusters with less than four interneurons were grouped into the gray “other” category. The clustering distribution for the k-means clustering is shown on the right, in the same color scheme used throughout the rest of the chapter. For each distribution, red whiskers represent the PV+ interneurons. Black lines connect corresponding interneurons that were categorized differently in each distribution, therefore fewer lines indicate greater overlap between clustering methods. The two clustering methods showed 82% overlap, meaning 82% of interneurons were categorized within the same cluster. The number of interneurons in each cluster is noted beside each cluster, along with the percentage of that cluster which was classified into their corresponding cluster in the other clustering method. For example, cluster 1 in the hierarchical clustering method has 26 interneurons, 91% of which were also classified into cluster 1 in the k-means clustering method. B, K-means clustering was used to cluster all 106 interneurons using only one type of data: either anatomic or electrophysiological. Bi, Distribution plots for purely anatomic clustering and purely electrophysiological clustering are shown as in Aii. Both anatomic and electrophysiological clustering were matched to the combined, four-cluster k-means clustering distribution, as described in Materials and Methods. The overlap between purely anatomic and purely electrophysiological clustering was 58%, indicating that some, but not most, interneurons could be matched to different anatomic and electrophysiological profiles. Bii, The same anatomic and electrophysiological distributions as in Bi are shown in comparison with the combined distribution in the center. There is a 70% overlap between the combined distribution and the anatomic, whereas there is 83% overlap between the combined and electrophysiological distribution.