Figure 2
From: New approach to determine the healthy immune variations by combining clustering methods
Clustering approaches applied to GEOCODE study cohort. (A) Three-dimensional PCA of the study was used as a qualitative visual criterion. (B) The graph shows the average silhouette indices obtained by running PAM with an increasing number of clusters K. For each value of K, 20 instances of the PAM algorithm were run with a random initialization of centroids, in order to mitigate the effect of local minima. The corresponding solutions are represented by black circles, whereas the best solutions for each K, i.e. the partitions which maximize the Silhouette Index, are connected by a plain red line. (C) The graph shows the Gap Statistic as a function of the number of clusters. For each value of K, the Gap Statistic was computed by comparing the clustering solution obtained on the dataset with partitions extracted from 200 unstructured datasets randomly generated. (D) The graph shows the results of the Cluster Stability Analysis, for which 100 random subsamples of the original dataset have been clustered and compared using the Jaccard Index. Graphs were performed using R Software version 3.6.1
