Challenges in structural variant calling in low-complexity regions (2509.23057v1)

Abstract: Background: Structural variants (SVs) are genomic differences $\ge$50 bp in length. They remain challenging to detect even with long sequence reads, and the sources of these difficulties are not well quantified. Results: We identified 35.4 Mb of low-complexity regions (LCRs) in GRCh38. Although these regions cover only 1.2% of the genome, they contain 69.1% of confident SVs in sample HG002. Across long-read SV callers, 77.3-91.3% of erroneous SV calls occur within LCRs, with error rates increasing with LCR length. Conclusion: SVs are enriched and difficult to call in LCRs. Special care need to be taken for calling and analyzing these variants.

• The paper quantifies that although low-complexity regions make up only 1.2% of GRCh38, they harbor over 69% of confident SVs, underscoring significant detection issues.
• It benchmarks 11 long-read SV callers against updated GIAB standards, revealing error rates of up to 91.3% in LCRs, especially for SV events over 2 kb.
• The study recommends advanced haplotype-aware multi-sequence realignment and pangenome-based approaches to improve SV calling accuracy in challenging genomic regions.

Structural Variant Calling in Low-Complexity Regions: Quantitative Challenges and Implications

Introduction

This paper provides a systematic quantification of the challenges associated with structural variant (SV) calling in low-complexity regions (LCRs) of the human genome. SVs, defined as genomic differences of at least 50 bp, are known to have significant functional impacts. Despite advances in long-read sequencing technologies and assembly-based approaches, accurate detection of SVs in LCRs remains problematic. The authors present a comprehensive analysis of LCRs in GRCh38 and T2T-CHM13, their overlap with SVs, and the error profiles of state-of-the-art SV callers, highlighting the disproportionate enrichment of SVs and errors within LCRs.

Identification and Annotation of Low-Complexity Regions

The paper employs longdust to identify LCRs in GRCh38, filtering out centromeric repeats using dna-brnn, resulting in 34.4 Mb of LCRs (≥50 bp). To account for polymorphic LCRs absent from GRCh38, the authors analyze 462 assemblies from the Human Pangenome Reference Consortium (HPRC), annotating variant bubbles in the minigraph graph. LCRs are prioritized below segmental duplications (SegDup) to avoid misclassification. The final merged annotation covers 35.4 Mb in GRCh38, with 16.2% intersecting SegDup regions. In T2T-CHM13, LCRs span 79.6 Mb, primarily due to additional centromeric satellites.

Benchmarking and SV Calling Methodology

The analysis leverages the latest GIAB SV benchmark (HG002-Q100 v1.1), which contains 28,188 SVs in 2.76 Gb of confident regions, contrasting with the older HG002-SV v0.6 benchmark (9,705 SVs in 2.66 Gb). The increased SV count in the newer benchmark is attributed to improved allele resolution and inclusion of LCRs. SVs are called using 11 long-read SV callers on PacBio HiFi data aligned to GRCh38, with accuracy evaluated against both benchmarks using truvari. Callers include cuteSV, DeBreak, Delly, longcallD, pbsv, Sawfish, Severus, Sniffles2, SVDSS, SVIM, and SVision-pro, with genotyping performed via kanpig for SVDSS calls.

Quantitative Enrichment and Error Profiles in LCRs

The results demonstrate that although LCRs constitute only 1.2% of GRCh38, they contain 69.1% of confident SVs in HG002. Across SV callers, 59.4–67.7% of SV calls overlap LCRs. Critically, 77.3–91.3% of erroneous SV calls are localized to LCRs, with error rates increasing as LCR length increases. SV callers that perform haplotype-resolved calling (e.g., longcallD, SVDSS) detect more SVs in LCRs and SegDup regions, but even these advanced methods struggle with long LCRs. The majority of errors are attributed to inconsistent read alignment in LCRs, particularly for events exceeding 2 kb, where some callers miss up to half of SVs despite sufficient read length.

Algorithmic Considerations and Practical Recommendations

The analysis reveals that simple SV calling algorithms lacking realignment or local reassembly are inadequate for LCRs. Haplotype-aware multi-sequence realignment, as implemented in longcallD, substantially reduces error rates. The authors recommend stratifying SVs by LCR for downstream analyses, given their distinct error profiles and biological relevance. Notably, LCRs may overlap coding exons and regulatory elements, precluding blanket filtering of LCR-overlapping SVs.

For high-coverage long-read datasets, haplotype-resolved assembly followed by assembly-to-reference alignment offers superior accuracy, as evidenced by the derivation of the HG002-Q100 truth set. However, merging SV calls across samples remains challenging, especially in LCRs. Pangenome-based approaches utilizing multi-sequence alignment across samples (e.g., minigraph-based methods) are preferable for consistent SV representation, though highly variable LCRs still pose difficulties.

Implications and Future Directions

The findings underscore the necessity for specialized algorithmic strategies in SV calling within LCRs. The disproportionate enrichment of SVs and errors in these regions has direct implications for population-scale variant discovery, clinical genomics, and functional genomics. Future developments should focus on improving realignment and assembly algorithms, integrating pangenome graphs, and refining benchmarking standards to better capture the complexity of LCRs. The expansion of reference genomes (e.g., T2T-CHM13) and increased sample diversity will further elucidate the landscape of LCRs and their impact on SV calling.

Conclusion

This paper provides a rigorous quantification of the challenges in SV calling within low-complexity regions, demonstrating that LCRs are both highly enriched for SVs and disproportionately responsible for calling errors. Advanced haplotype-aware algorithms and assembly-based approaches are essential for accurate SV detection in these regions. The results have significant implications for the design of SV calling pipelines, benchmarking standards, and the interpretation of SVs in functional and clinical genomics. Continued methodological innovation and comprehensive reference resources will be critical for addressing the persistent challenges posed by LCRs in SV analysis.