Characterizing and measuring bias in sequence data.

Diagram illustrating the low coverage of NCS1 exon 1 in 198× Illumina HiSeq shotgun data. The first 72 bases of the first exon of human gene NCS1, including the transcription start site, were uncovered in a 198× whole-genome shotgun data set (#A2). The displayed 2,000 base region is chromosome 9:132,933,910-132,935,910. NCS1 encodes calcium-binding proteins that regulate neurotransmitter release [1].

GC-bias plots for three microbial genomes. Top: plots showing the relative coverage GC-bias for Illumina MiSeq, Ion Torrent PGM, and Pacific Biosciences RS on the P. falciparum (19% GC), E. coli (51%), and R. sphaeroides (69%) genomes (Table 2, data sets 1 to 9). Unbiased coverage would be represented by a horizontal line at a relative coverage = 1 (black dashed line). Relative coverage is only plotted for GC percentages for which there are at least 1,000 100-base windows in the genome. Bottom: the GC composition distribution of each genome.

GC-bias plots for the human genome. Left: the GC composition distribution of the human genome (HG19, GRCh37). Center and right: GC-bias plots for several data sets from human NA12878. Unbiased coverage would be represented by a horizontal line at relative coverage = 1. Center: HiSeq v3 with sample-preparation reagents from Kapa Biosystems (Table 2, data set 14), Ion Torrent PGM (data set 15), and Complete Genomics data (data set 16). Right: HiSeq v3 with sample-preparation reagents from Kapa Biosystems (data set 14, as in center panel) and HiSeq v3 with the standard Fisher et al. [31] reagents (data set 13). Note that Illumina relative coverage exceeded the y-axis above 93% GC content. Relative coverage is only plotted for GC percentages for which there are at least 1,000 100-base windows in the genome.

Error rates as a function of GC composition. Each graph shows mismatch (light blue), deletion (dark blue), and insertion (maroon) rates (y-axis) as a function of GC composition (x-axis). Data are shown for the Ion Torrent PGM from three organisms (P. falciparum, R. sphaeroides, and human), for the Illumina MiSeq on the two microbes, for the Illumina HiSeq on human, for Pacific Biosciences from the two microbes and from Complete Genomics for human (Table 2, data sets 1 to 3, 7 to 9, and 14 to 16). For human we note that bona fide differences between the sample and the reference sequence were recorded as errors. Error rates are only plotted for GC percentages for which there are at least 1,000 100-base windows in the genome.

Error rates as a function of homopolymer length. Each graph shows mismatch (light blue), deletion (dark blue), and insertion (maroon) rates (y-axis) within homopolymers of various lengths (x-axis). Data are plotted from P. falciparum and human as available (Table 2, data sets 1 to 3 and 14 to 16). For human we note that bona fide differences between the sample and the reference sequence were recorded as errors.

GC and homopolymer distributions of uncharacterized Illumina undercoverage of human sample NA12878. The graphs show the distribution of GC-content and homopolymer length for bases in the overall human genome and in the genome intervals that are ten-fold undercovered but which were not explained by known sequence biases or differences between the sample and reference sequence. Data are from Table 2, data set 14.