Fig. 1From: HycDemux: a hybrid unsupervised approach for accurate barcoded sample demultiplexing in nanopore sequencingAn example of clustering 30 sequences using our hybrid clustering algorithm (A, B, and C) and the subsequent demultiplexing mechanism based on the clustering results (D). A We perform initial clustering on 30 sequences, resulting in 6 clusters. If a cluster contains a number of sequences greater than the GoodIndex, it is considered a good cluster. In this case, clusters \(C_1\), \(C_2\), and \(C_3\) are considered good clusters. B We attempt to merge the good clusters by selecting k signals from each cluster. If the distance values in the corresponding \(k \times k\)Â dynamic time warping (DTW) distance matrix are all smaller than a given threshold, we merge two clusters. In this case, \(C_1\) and \(C_2\) are merged into a single cluster. C We attempt to add sequences that are not in good clusters to the existing good clusters after the cluster merging is completed. We select a representative sequence from each good cluster and calculate the DTW distance between the representative sequence and the sequences not in the good cluster. If the distance value is less than a given threshold, we add the sequence to the corresponding good cluster. D We demultiplex the cluster by selecting k signals within the cluster and converting all known barcode sequences into standard nanopore signals based on the official 6-mer table of Oxford Nanopore Sequencing Company. We then calculate the DTW distance matrix between these k signals and the standard nanopore signals, and find the row label corresponding to the minimum value of each column to obtain a one-dimensional row label matrix. Finally, we compute the mode of the row-labeled matrix and use it as the final demultiplexed resultBack to article page