Fig. 7

Framework of seed selection model. Assume that a set of M single-copy marker genes are mapped to 15 contigs with hidden origins indicated by different colors, we first estimate the number of genomes as G = 5 and determine a set (\(\mathcal{S}\)) of groups of contigs containing marker genes {m₄, …, m_M}. Then the classifier is trained with each group of contigs from \(\mathcal{S}\) and predicts all contigs in \(\mathcal{S}\) based on both k-mer frequencies and coverages to form a symmetric similarity matrix. We next run spectral clustering algorithm to obtain spectral clusters. And finally, contigs in each spectral cluster are concatenated into a single extended seed