Fig. 3

Simulating topological complexity and missing data. Four plots measuring the Jaccard Index (see In test) against increasing topological complexity and across increasing levels of missing data percentage (colored as indicated in the insert) are shown. Each curve in each plot, therefore, corresponds to a specific missing data condition. We compared four methods: (a) Clusterflock, (b) multidimensional scaling (MDS), (c) hierarchical clustering, and (d) partitioning around medoids (PAM). To compare the four techniques using similar methods, because the number of underlying topologies in each simulation was known, for Clusterflock and MDS, we used k-means to cluster the final spatial arrangement of loci and assign OGFs to topological groups