Characterization and mitigation of artifacts derived from NGS library preparation due to structure-specific sequences in the human genome.

backgroundHybridization capture-based targeted next generation sequencing (NGS) is gaining importance in routine cancer clinical practice. DNA library preparation is a fundamental step to produce high-quality sequencing data. Numerous unexpected, low variant allele frequency calls were observed in libraries using sonication fragmentation and enzymatic fragmentation. In this study, we investigated the characteristics of the artifact reads induced by sonication and enzymatic fragmentation. We also developed a bioinformatic algorithm to filter these sequencing errors.

resultsWe used pairwise comparisons of somatic single nucleotide variants (SNVs) and insertions and deletions (indels) of the same tumor DNA samples prepared using both ultrasonic and enzymatic fragmentation protocols. Our analysis revealed that the number of artifact variants was significantly greater in the samples generated using enzymatic fragmentation than using sonication. Most of the artifacts derived from the sonication-treated libraries were chimeric artifact reads containing both cis- and trans-inverted repeat sequences of the genomic DNA. In contrast, chimeric artifact reads of endonuclease-treated libraries contained palindromic sequences with mismatched bases. Based on these distinctive features, we proposed a mechanistic hypothesis model, PDSM (pairing of partial single strands derived from a similar molecule), by which these sequencing errors derive from ultrasonication and enzymatic fragmentation library preparation. We developed a bioinformatic algorithm to generate a custom mutation "blacklist" in the BED region to reduce errors in downstream analyses.

conclusionsWe first proposed a mechanistic hypothesis model (PDSM) of sequencing errors caused by specific structures of inverted repeat sequences and palindromic sequences in the natural genome. This new hypothesis predicts the existence of chimeric reads that could not be explained by previous models, and provides a new direction for further improving NGS analysis accuracy. A bioinformatic algorithm, ArtifactsFinder, was developed and used to reduce the sequencing errors in libraries produced using sonication and enzymatic fragmentation.