TL;DR
This paper introduces OMSV, a novel computational pipeline for accurately identifying large structural variations (SVs) in the human genome using single-molecule optical mapping. OMSV significantly outperforms existing methods in both sensitivity and specificity, and importantly, can detect SVs missed by traditional sequencing approaches, with experimental validation confirming its accuracy.
What Problem Does This Paper Address?
Detecting structural variations (SVs) in the human genome is crucial for understanding disease and phenotypic diversity. While sequencing technologies have advanced, their limited read lengths make it challenging to accurately identify large SVs, especially those involving repetitive regions or novel sequences. Nanochannel-based optical mapping (OM) offers a solution by allowing direct visualization of very long DNA molecules (up to megabases), making it ideal for detecting large SVs. However, OM data is prone to various errors (e.g., false positives/negatives, sizing errors), and existing methods struggle to simultaneously handle these errors, detect all SV types, and distinguish between homozygous and heterozygous variants. This paper addresses the need for a robust, accurate, and comprehensive SV detection tool for optical mapping data.
Methods
The authors developed OMSV, a two-step pipeline:
- Alignment: Optical maps are aligned to a reference map derived from the human reference sequence. This step uses two different aligners, RefAligner (for highly similar maps) and OMBlast (for more complex rearrangements), and their results are integrated to create consensus alignments.
- SV Calling: The consensus alignments are fed into three specialized modules:
- Extra/Missing Restriction Sites Caller: Identifies SVs involving the creation or removal of restriction sites, handling both homozygous and heterozygous variants.
- Large Size Change Caller: Detects SVs characterized by significant changes in the distance between restriction sites, even if no sites are gained or lost.
- Complex SV Caller: Identifies inversions, translocations, and copy number variations (CNVs) using split-alignment strategies and depth-of-coverage analysis.
A formal error model is integrated throughout the pipeline to account for the inherent noise in optical mapping data and to evaluate the likelihood of different SV hypotheses.
Datasets
The authors used two primary types of datasets for evaluation:
- Simulated Data: They generated simulated optical mapping data from artificial haploid and diploid human genomes. These datasets were created by introducing various types of genetic variations (SNVs, indels, complex SVs) to the human reference genome (hg38) and then simulating noisy optical maps based on a comprehensive error model. Multiple simulation sets were generated by varying error rates and coverage depths.
- Real Data:
- Family Trio: Optical maps from a previously published family trio (NA12878, NA12891, NA12892) were used.
- Human Cell Line (C666-1): Optical maps from the C666-1 cell line were generated and analyzed.
Key Insights and Ideas
- OMSV's Robustness: The method is designed to be robust against the inherent errors in optical mapping data, employing a formal error model and integrating results from multiple aligners.
- Comprehensive SV Detection: OMSV can detect a wide spectrum of SVs, including insertions, deletions, inversions, translocations, and CNVs, and importantly, can distinguish between homozygous and heterozygous variants.
- Bridging the Gap with Sequencing: A significant insight is that OMSV can identify large SVs that are often missed by short-read sequencing-based callers, highlighting the complementary nature of these technologies.
- Leveraging Sequencing for Precision: The paper demonstrates how OMSV can be combined with sequencing data to precisely determine SV break points and even infer novel sequences within SVs not present in the reference.
- Open-Source Availability: The authors provide OMSV as open-source software, which is crucial for widespread adoption and further research in the field.
Key Results
- High Accuracy on Simulated Data: OMSV achieved high precision (90-99%) and recall (81-98%) for indels larger than 2kbp on simulated data, outperforming the existing BioNano Solve SV caller in recall.
- Complex SV Detection: OMSV showed good precision (80-85%) for complex SVs, with a recall rate of 45-80% after filtering for intrinsically feasible cases. BioNano Solve failed to detect any complex SVs from the simulated data.
- Real Data Validation: In a family trio, OMSV identified hundreds of SVs, with high Mendelian concordance (98% ignoring zygosity, 85% considering zygosity).
- Missed SVs by Sequencing: When applied to the C666-1 cell line, OMSV identified hundreds of SVs >2kbp, with 65% of them missed by sequencing-based callers (including 80% of insertions).
- Experimental Validation: Seven selected insertions identified by OMSV were experimentally validated using PCR, confirming their presence.
- Speed: OMSV's running time is significantly faster than assembly-based methods, with alignments being the main bottleneck, which can be parallelized.
Limitations and Future Work
- Complex SV Recall: While OMSV shows good precision for complex SVs, the recall rate is moderate (30-50% initially, improving to 45-80% for feasible cases). Some SVs, like inversions without nicking sites, are intrinsically difficult to call.
- Zygosity Determination in Real Data: The paper notes that while OMSV accurately identifies SV locations, determining the correct zygosity of SVs can be more challenging with real data compared to simulated data.
- Alignment Bottleneck: The primary limitation in terms of speed is the optical map alignment step, although this can be mitigated by parallel processing.
- Future Work: The authors suggest that OMSV can be extended to study haplotypes, cell type composition, and cell-to-cell variability, which would require highly accurate alignment of individual labels. Probing nicking sites with a second enzyme could further improve alignment accuracy. The method's ability to incorporate sequencing data for precise break points and novel sequence identification is also highlighted as a key future direction.
Summary updated