Assembling single cell reads using Velvet-SC. (a) Coverage varies widely along the genome, between 1 and 12 in this cartoon example. Reads (short lines) and potential contigs (thick lines; boxes around the supporting reads) are positioned along the genome, with a box around the reads supporting each contig. There are two potential contigs to choose from in the middle, differing by a single nucleotide (C vs. T): a green contig with coverage 6.4, and a blue contig with coverage 1. With a fixed coverage threshold of 4, Velvet would delete the low coverage blue and purple contigs, and then merge the high coverage red and green contigs into a contig much shorter than the full genome. Velvet-SC instead starts by eliminating sequences of average coverage 1, which only removes the blue contig. The other contigs are combined into a single contig (b) of average coverage 9. The purple region is salvaged by Velvet-SC because it was absorbed into a higher coverage region coverage threshold increased. Velvet-SC repeats this process with a gradually increasing low coverage threshold. (c) A portion of the de Bruijn graph for the contigs described in (a). The black circles are the “vertices” and represent 5-mer strings derived from the reads, which are indicated by colored lines alongside the chains of vertices, including a blue read with an erroneous T. The lines between the vertices are termed “edges” and represent the overlaps between the 5-mers. The edges are directed from left to right in this example. The read with the C/T mismatch results in two alternative paths for assembly, both with 5 intermediate vertices. The lower of the two paths arises from the erroneous blue read and has coverage 1; it is the only part of the graph eliminated by Velvet-SC, leaving a single chain of vertices that gives a single contig for the entire genome. See for an example of the condensing of contigs. An example of Velvet-SC handling of a chimeric read is presented in .