Fig. 3
The steps of the QFMatch algorithm as applied in aligning clusters identified by the EPP approach. a–c. Merge the previously clustered samples (panel a, samples were clustered as described in Fig. 2b) and perform adaptive binning (panel b) insuring 2lnN events per bin, where N is the number of events in the smallest sample. Split the samples back while preserving the binning pattern (panel c). Calculate quadratic form dissimilarity6 between each possible combination of cluster pairs (Groups #1-#4 from Sample A and Sample B), on which medians are located no more than four standard deviations apart in every dimension (Table 1). Pairs with the smallest dissimilarity scores are considered as matched. The remaining clusters in each sample are automatically treated as merging candidates (Table 2). If there is more than one merging candidate then all possible permutations of merging candidates are considered.
