Family reunion via error correction: An efficient analysis of duplex sequencing data

Abstract

Duplex sequencing is the most accurate approach for identification of sequence variants present at very low frequencies. Its power comes from pooling together multiple descendants of both strands of original DNA molecules, which allows distinguishing true nucleotide substitutions from PCR amplification and sequencing artifacts. This strategy comes at a cost-sequencing the same molecule multiple times increases dynamic range but significantly diminishes coverage, making whole genome duplex sequencing prohibitively expensive.Furthermore, every duplex experiment produces a substantial proportion of singleton reads that cannot be used in the analysis and are, technically, thrown away. In this paper we demonstrate that a significant fraction of these reads contains PCR or sequencing errors within duplex tags. Correction of such errors allows "reuniting" these reads with their respective families increasing the output of the method and making it more cost effective.Additionally, we combine error correction strategy with a number of algorithmic improvements in a new version of the duplex analysis software, Du Novo 2.0, readily available through Galaxy, Bioconda, and as the source code.

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

“…These tags with errors will be effectively lost if not re-united with a barcode correction tool to the original family (Stoler et al 2018).…”

“…The tag analysis distance was estimated as reported in Stoler et al (Stoler et al 2018). We used the same list of tags as already described in the FSD.…”

“…However, sequencing or PCR errors could introduce 1-2 nucleotide differences in the tags that would be otherwise identical. These tags with errors will be effectively lost if not re-united with a barcode correction tool to the original family (Stoler et al 2018).…”

“…This is verified in a TD analysis using only tags forming DCS, which should be error-free ( Fig 2B and Fig S3B show that almost no tags with TD=1 are found in the DCS population). Thus, in these data sets it is appropriate to implement the barcode correction tool (Stoler et al 2018) allowing for at least one mismatch in the tags, which improves the duplex coverage from 23% to 32% and 18% to 23% in the PF2-CRISPR and PF2 -Standard libraries, respectively. As expected, the same analysis of TD after the barcode correction with one mismatch now renders mainly tags with a TD = 5-7, since tags with one difference have been re-united with larger families (thus, TD =1 is zero).…”

“…Changes to the library preparation, such as CRISPR/Cas targeted digestion, reduces the number of amplification steps and increases the number of DCS per input material (6-12%) (Nachmanson et al 2018). Also changes in the bioinformatic pipeline have improved the data output such as the Du Novo reference-free consensus building (Stoler et al 2016) and error correction in the tags to re-unite otherwise "lost" reads into their respective families (Stoler et al 2018).…”