Expectations and blind spots for structural variation detection from short-read alignment and long-read assembly

Xuefang Zhao, Ryan L. Collins, Wan-Ping Lee, Alexandra M. Weber, Yukyung Jun, Qihui Zhu, Ben Weisburd, Yongqing Huang, Peter A. Audano, Harold Wang, Mark Walker, Chelsea Lowther, Jack Fu, Mark Gerstein, Scott E. Devine, Tobias Marschall, Jan O. Korbel, Evan E. Eichler, Mark Chaisson, Charles Lee, Ryan E. Mills, Harrison Brand, Michael E. TalkowskibioRxiv 2020 · DOI

TL;DR

This paper benchmarks short-read (srWGS) and long-read (lrWGS) sequencing technologies for detecting structural variants (SVs), revealing that while srWGS captures most deletions in well-behaved genomic regions, lrWGS is crucial for uncovering complex insertions and variants in challenging repetitive regions. These findings help set expectations for current srWGS studies and highlight the future value of lrWGS for discovering novel disease-associated variants.

What Problem Does This Paper Address?

The field of genomics heavily relies on short-read whole-genome sequencing (srWGS) for large-scale studies like biobanks and disease diagnostics. However, srWGS has inherent limitations in accurately detecting structural variants (SVs), which are large-scale genomic alterations. While long-read whole-genome sequencing (lrWGS) technologies offer superior capabilities for SV detection and genome assembly, they are currently cost-prohibitive for widespread use, and far fewer lrWGS genomes are publicly available compared to the millions of srWGS genomes. This disparity creates a gap in our understanding of SVs, particularly in complex or repetitive genomic regions. The paper aims to bridge this gap by quantifying the SV detection capabilities of both srWGS and lrWGS, establishing benchmarks for current srWGS studies, and projecting the added value of lrWGS for future discoveries.

Methods

The researchers compared SV detection from srWGS and lrWGS using matched samples from three parent-child trios within the Human Genome Structural Variation Consortium (HGSVC). For srWGS, they employed an ensemble approach using 13 SV detection algorithms. For lrWGS, they utilized haplotype-resolved assembly from PacBio sequencing. To address discordance and improve precision, they developed an in silico SV assessment procedure. This procedure re-evaluated SVs using three orthogonal pieces of information: evidence from the VaPoR algorithm (which surveys raw lrWGS reads), copy states based on srWGS normalized read depth, and discordant paired-end and split reads at SV breakpoints. SVs supported by one or more of these evidence classes were considered "high confidence." They also analyzed the alignment of assembled lrWGS insertion sequences against a catalog of known repeat elements.

Datasets

The primary dataset used for evaluation consisted of three parent-child trios from the Human Genome Structural Variation Consortium (HGSVC). These trios had matched srWGS and lrWGS data available. The paper references existing genomic annotations and repeat element catalogs (e.g., RepeatMasker, UCSC genome browser) for genomic context analysis.

Key Insights and Ideas

  • Genomic context is king: The detectability and concordance of SVs are dramatically influenced by the genomic region. Highly repetitive regions (segmental duplications and simple repeats) pose significant challenges for srWGS and show lower concordance between srWGS and lrWGS.
  • lrWGS is a game-changer for repetitive regions: While srWGS can capture most deletions in "unique" (non-repetitive) genomic sequences, lrWGS is essential for discovering variants, especially deletions, in repetitive regions that are often missed by srWGS.
  • Insertions are a blind spot for srWGS: lrWGS significantly outperforms srWGS in detecting insertions, regardless of genomic context. This is particularly important for identifying functional insertions and transposable elements.
  • srWGS is not entirely useless for insertions: Even for lrWGS-identified insertions not detected by srWGS, there is often some detectable support in the raw srWGS data, suggesting that improved srWGS algorithms could enhance insertion detection.
  • The "unsolved" cases: While lrWGS might not dramatically increase the discovery of simple deletions in well-annotated coding regions (where srWGS already performs well), it holds immense promise for uncovering novel disease-associated variants in previously inaccessible genomic regions, particularly those involving repeat expansions and mobile elements.

Key Results

  • SV Yield: srWGS captured approximately 11,000 SVs per genome, while haplotype-resolved assembly from lrWGS identified approximately 25,000 SVs per genome.
  • Deletion Concordance: In non-repetitive regions (90.3% of the human reference), high concordance (93.8%) was observed for deletions between srWGS and lrWGS after error correction.
  • Repetitive Region Impact: Segmental duplications (SD) and simple repeats (SR) constitute 9.7% of the reference genome but contained breakpoints for a substantial proportion of SVs (49.8% for srWGS, 70.4% for lrWGS). Concordance in these regions was significantly lower (57.0% for srWGS, 22.5% for lrWGS).
  • lrWGS Superiority for Insertions: Nearly half (46.3%) of high-confidence lrWGS insertions in unique regions had no matching SV call from srWGS, while the majority (94.7%) of srWGS insertions were captured by lrWGS.
  • Repeat Element Alignment: A significant portion of lrWGS-identified insertions aligned to specific types of repeat elements (e.g., SINEs, LINEs, LTRs), highlighting the role of these elements in SV formation.

Limitations and Future Work

  • Cost of lrWGS: The paper acknowledges that the ~5-12X higher cost of lrWGS compared to srWGS makes it impractical for large-scale population studies currently.
  • Low Throughput of lrWGS: The low throughput of current lrWGS platforms also limits their immediate widespread adoption.
  • Technical Noise: The study highlights that a significant portion of discordance between srWGS and lrWGS, particularly for deletions in non-repetitive regions, is likely due to technical noise and false positive calls, which can be mitigated by post hoc filtering.
  • Future Directions: The authors suggest that continued development of detection algorithms could improve srWGS sensitivity for missed insertion SVs. They also emphasize the long-term value of assembly-based lrWGS for creating new catalogs of functional insertions, transposable elements, and disease-associated repeat expansions in previously recalcitrant genomic regions. As telomere-to-telomere assembly methods mature, they will further expand the catalog of disease-associated variants.

Summary updated

Abstract

Abstract Virtually all genome sequencing efforts in national biobanks, complex and Mendelian disease programs, and emerging clinical diagnostic approaches utilize short-reads (srWGS), which present constraints for genome-wide discovery of structural variants (SVs). Alternative long-read single molecule technologies (lrWGS) offer significant advantages for genome assembly and SV detection, while these technologies are currently cost prohibitive for large-scale disease studies and clinical diagnostics (∼5-12X higher cost than comparable coverage srWGS). Moreover, only dozens of such genomes are currently publicly accessible by comparison to millions of srWGS genomes that have been commissioned for international initiatives. Given this ubiquitous reliance on srWGS in human genetics and genomics, we sought to characterize and quantify the properties of SVs accessible to both srWGS and lrWGS to establish benchmarks and expectations in ongoing medical and population genetic studies, and to project the added value of SVs uniquely accessible to each technology. In analyses of three trios with matched srWGS and lrWGS from the Human Genome Structural Variation Consortium (HGSVC), srWGS captured ∼11,000 SVs per genome using reference-based algorithms, while haplotype-resolved assembly from lrWGS identified ∼25,000 SVs per genome. Detection power and precision for SV discovery varied dramatically by genomic context and variant class: 9.7% of the current GRCh38 reference is defined by segmental duplications (SD) and simple repeats (SR), yet 91.4% of deletions that were specifically discovered by lrWGS localized to these regions. Across the remaining 90.3% of the human reference, we observed extremely high concordance (93.8%) for deletions discovered by srWGS and lrWGS after error correction using the raw lrWGS reads. Conversely, lrWGS was superior for detection of insertions across all genomic contexts. Given that the non-SD/SR sequences span 90.3% of the GRCh38 reference, and encompass 95.9% of coding exons in currently annotated disease associated genes, improved sensitivity from lrWGS to discover novel and interpretable pathogenic deletions not already accessible to srWGS is likely to be incremental. However, these analyses highlight the added value of assembly-based lrWGS to create new catalogues of functional insertions and transposable elements, as well as disease associated repeat expansions in genomic regions previously recalcitrant to routine assessment.
BibTeX
@misc{zhao2020expectations,
  author = {Xuefang Zhao and Ryan L. Collins and Wan-Ping Lee and Alexandra M. Weber and Yukyung Jun and Qihui Zhu and Ben Weisburd and Yongqing Huang and Peter A. Audano and Harold Wang and Mark Walker and Chelsea Lowther and Jack Fu and Mark Gerstein and Scott E. Devine and Tobias Marschall and Jan O. Korbel and Evan E. Eichler and Mark Chaisson and Charles Lee and Ryan E. Mills and Harrison Brand and Michael E. Talkowski},
  title = {{Expectations and blind spots for structural variation detection from short-read alignment and long-read assembly}},
  year = {2020},
  doi = {10.1101/2020.07.03.168831},
  howpublished = {bioRxiv},
}
Figure