Increased yields of duplex sequencing data by a series of quality control tools

Abstract

Duplex sequencing is currently the most reliable method to identify ultra-low frequency DNA variants by grouping sequence reads derived from the same DNA molecule into families with information on the forward and reverse strand. However, only a small proportion of reads are assembled into duplex consensus sequences, and reads with potentially valuable information are discarded at different steps of the bioinformatics pipeline, especially reads without a family. We developed a bioinformatics tool-set that analyses the tag and family composition with the purpose to understand data loss and implement modifications to maximize the data output for the variant calling. Specifically, our tools show that tags contain PCR and sequencing errors that contribute to data loss and lower DCS yields. Our tools also identified chimeras, which result in unpaired families that do not form DCS. Finally, we also developed a tool called Variant Analyzer that re-examines variant calls from raw reads and provides different summary data that categorizes the confidence level of a variant call by a tier-based system. We demonstrate that this tool identified false positive variants tagged by the tier-based classification. Furthermore, with this tool we can include reads without a family and check the reliability of the call, which increases substantially the sequencing depth for variant calling, a particular important advantage for low-input samples or low-coverage regions.

Discovery of an unusual high number ofde novomutations in sperm of older men using duplex sequencing

“…Seven libraries featured younger donors with ages ranging from 26 to 30 years, and the remaining six featured older donors with ages ranging from 49 to 59 years; see Supplemental Table S3 and Supplemental Table S4. Consensus sequences were first formed using Du Novo (Stoler et al 2016;Stoler et al 2020) and variants were called using the Naive Variant Caller (Blankenberg et al 2014), followed by a more thorough screening with our Variant Analyzer software (Povysil et al 2021). The Variant Analyzer re-examines raw reads from Du…”

“…With this tier classification (see Supplemental Table S5), we can distinguish true mutants from bad quality data or biases leading likely to false positives. Moreover, our Variant Analyzer supports the analysis of reads without a family as demonstrated in (Povysil et al 2021), which allowed maximizing the data output.…”

“…according to a DS specific pipeline that includes an error correction tool (Stoler et al 2020). Variants (substitutions only) were further inspected and assigned to tiers using the Variant Analyzer (Povysil et al 2021). Every variant was categorized as a SNP if detected in all libraries that shared the same donor pool/target region combination with a frequency ~10%.…”

“…Variants with DCS coverage below 1000, intronic variants and SNPs were discarded from our analysis and only exonic variants of Tier 1 were kept, together with Tier 2 that were detected more than once. For more details on this analysis see (Povysil et al 2021). Selected variants were annotated using Variant Effect Predictor (VEP) (McLaren et al 2016) and wANNOVAR (Yang and Wang 2015).…”