Schematic diagram of the sequencing strategy. Sample pools of 40 samples × 12 pools were generated from a cohort of 480 individuals for PCR amplification of individual exons. After blunt-ended ligation and random fragmentation, PCR amplicons from individual sample pools were used to generate indexed sequence libraries. The 12 indexed libraries were combined in equal molar amounts and sequenced in one lane of a flow cell using an Illumina GAII.

Quality assessment of the Illumina sequence data. (a) Number of reads with barcodes that passed Illumina filtering and aligned to the reference templates using Bowtie from individually indexed libraries (n = 12). Range, 641 k to 978 k reads; mean ± standard deviation, 809 k ± 107 k. (b) Percentage of total (unaligned) reads that fall into a mean Phred quality interval. Note > 80% of the reads have mean Phred quality scores ≥25. (c) Nucleotide content as a function of sequencing cycles (n = 47). Note that the nucleotide proportions closely match the expected proportions as determined from the templates.

Representative base reads and tailcurves for common and rare variants and error calls. (a) Position with no variant. (b) Position with a common variant. (c) Position with a rare variant. (d) Position with a false positive call.

Continuity versus weighted allele frequency curves for select variants. (a) Very common variant present in all 12 pools. (b) Modestly common variant present in the majority of pools. (c) Infrequent variant present in a minority of pools. (d) Rare variant present in only one pool. Gold circles indicate variant pools retained by cluster analysis, while a gold 'x' indicates a variant pool that has been eliminated.

Amplicon ligation, fragmentation and indexed Illumina libraries. (a) Amplicon ligation and fragmentation: L-1, low molecular weight marker; lane 1, PCR amplicons before ligation; lane 2, PCR amplicons after ligation; lane 3, random fragmentation using Fragmentase (NEB). ^#The bracket indicates fragments of desired length. (b) Indexed Illumina libraries: L-2, 1-kb ladder; lanes 1 to 12, size distribution of 12 indexed Illumina libraries.

Distribution of quality score from SAMtools Pileup. Filtering was conducted at the 95th percentile of the consensus and SNP quality distributions reported by SAMtools; only the distribution of SNP quality values is depicted here. The blue bar is the 95th percentile score cutoff, discounting variants with max score. (a) SNP quality scores derived from Illumina base calls. (b) SNP quality scores derived from Srfim base calls.

Local pool patterns for error analysis. X-axes denote position in a local sequence. Position 16 is the variant site being analyzed, positions 1 to 15 are immediately upstream and positions 17 to 31 are downstream. Y-axes denote the weighted allele frequency of the most prominent non-reference allele at each position (mismatch rate). Individual pools are denoted by a unique line pattern, color, and number/letter. Light shading indicates the pool pattern that is most recognizable by SERVIC⁴E for each position. (a) Local weighted allele frequencies for each pool at position 14,551,524 ± 15 in chromosome 3 from the first cohort. The evaluated pattern of pools at the variant position involves pools 5, 6, 7, and 8, while the evaluated pattern at proximal positions involves pool 4. The dissimilarity between patterns results in retention of chr3:14551524 as a variant site. (b) Local weighted allele frequencies for each pool at position 14,552,916 ± 15 in chromosome 3 from the second cohort. The evaluated pattern of pools at the variant position involves pools 7, 13 (c), 20 (j), 22 (l), and 24 (n), and the evaluated pattern at proximal positions involves the same pools. The similarity between patterns results in elimination of chr3:14552916 as a variant site.

Average quality versus weighted allele frequency for variant pools after filtering by clustering. The X-axis is average Phred sequencing quality score and the Y-axis is weighted allele frequency (ratio of the sum of Phred quality scores for the variant allele at a position to the sum of all Phred quality scores at that position) in log₁₀ scale. Characteristic distribution shapes make it possible to cluster and retain only high quality variants (orange points). (a) Illumina base calls. (b) Srfim base calls.