Fig. 3

Extraction of a pathogenic Escherichia coli MAG from several closely related organisms in the metagenome from the gut of a travelers’ diarrhea patient (sample no. 76, with 2.92-Gbp raw reads, 10,910 contigs totaling 69.5 Mbp). Marker size (radius) is proportional to the cube root of contig length. Marker opacity is proportional to the cube root of contig coverage. Colors were assigned to the most abundant taxa in the sample. The assembly data (A, B) was explored using alternative dimensionality reduction methods (t-SNE for A, C, E–H; UMAP for B, D, both based on k-mer (k = 6) frequencies). A distinct blob of Enterobacteriaceae contigs (A, dashed line) were selected (C, D) and filtered based on its coverage profile (C, inset), resulting in a putative E. coli bin (E), which was further filtered by taxonomy (exemplified by arrows in E) to improve purity (F). In parallel, the corresponding bin inferred by MaxBin (G) was filtered by spatial pattern and taxonomy to retain a purer bin (H)