Fig. 6

The workflow of hybrid clustering integrated in HycDemux. First, a greedy clustering algorithm is used to obtain initial clustering based on the nucleotide information. Then, we try to merge the initial clusters with good homogeneity by the information of raw signals, in which the threshold of signal-to-signal DTW distance is determined with a partial sampling method. For each merged cluster, a corresponding representative signal is produced. Finally, based on these representative signals, a signal-similarity based classification strategy is designed to assign the unclassified sequences to the already known clusters, to further refine the clustering result