Automated, scalable quality control of heterogeneous exome sequence data
Year: 2016
Type: other
Abstract: In the coming years, hundreds of thousands of samples will be sequenced to identify associations between genetic variation and complex phenotypes. To avoid false-positive findings but maintain power, stringent quality control (QC) procedures must be applied to these data. Current QC approaches require the direct attention of analysts, presenting scaling challenges to millions of samples. We developed a novel approach for sample QC, combining control for ancestry and other confounders with automated outlier detection. Our approach (a) projects genotypes onto principal components (PCs) of known reference populations; (b) computes a battery of metrics assessing sequence quality; (c) normalizes metrics according to ancestry and other confounders such as sequencing platform; (d) performs PCA on the normalized metrics to identify independent measures of sample quality; and (e) uses Gaussian Mixture Modeling to identify outliers according to both the individual metrics as well as their PCs. We evaluated our method relative to previously published manual approaches using 18 sample metrics computed across 12 ancestry groups from 26,171 exomes. Sample quality was evaluated based on the transition-transversion ratio (TITV) of singleton variants unique to samples that passed QC (QC+) relative to those that failed (QC-). We identified 336 QC- samples, within which singleton variants had a TITV ratio of 1.75 compared to a ratio of 2.66 for QC+ samples (p=1.63e-129), suggesting QC- samples as significantly enriched for sequencing artifacts. Among them, 210 problematic samples with a singleton TITV ratio of 1.81 (p=2.8e-76) were not identified by the current gold standard approach. Samples substantially skewed on any of the metrics, including low call rate, high number of alternate alleles, and fraction of alternate reads at heterozygous sites (allele balance), were enriched for sequencing artifacts. Metrics based on lower frequency variants were more effective than comparable metrics based on common variants. PCA of normalized metrics identified six additional outliers with a mean singleton TITV ratio of 2 (p=6.8e-3), suggesting samples skewed across multiple metrics are also enriched for sequencing artifacts. These results demonstrate that not only can automatic, scalable QC approaches be developed for heterogeneous exome sequencing data, but their performance can meet or exceed that of current gold standard approaches.
Authors: Ryan Koesterer, Marcin von Grotthuss, Jason Flannick, GoT2D, T2D-GENES, SIGMA, LuCAMP, ESP
Institution: Broad Institute
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: gold
OpenAlex ID: W2561678603