Sequencing & Pangenomes

Improving structural variant discovery in low-depth Illumina sequencing using caller intersections

---

Abstract

Structural variants (SVs) are major drivers of phenotypic diversity and adaptation, but their accurate detection from short read Illumina data remains challenging, especially at low sequencing depth. Long read sequencing is the gold standard for SV discovery, yet it is not feasible at population scale. Consequently, most population scale studies rely on short read data, often at only ~10à— coverage, far below the 30à— depth used to train and benchmark common SV callers. This leads to high false positive rates and poor reproducibility, limiting SV discovery in livestock and other non model genomes. Although combining multiple callers is often used to boost recall, this strategy frequently increases false positives, as not all caller combinations are equally reliable. To address this, we developed a data driven filtering strategy that retains only those mutually exclusive caller intersection groups with proven precision benefit. Using 100 independent subsamples of the GIAB HG002 benchmark (90à— downsampled to ~10à—) and six SV callers (CNVnator, CUE, DELLY, GRIDSS, LUMPY, and Manta), we quantified each mutually exclusive caller intersection group's marginal precision contribution against the GIAB truth set. Only 13 groups had a positive precision impact and were retained. In the truth set, this selective filtering increased precision from 90.8% to 98.2% with minimal recall loss (11.2% → 10.4%), and SV reproducibility across subsamples increased significantly from 29.7% to 41.8% (p 10 kb), known to be error prone at low depth, were preferentially removed in both the truth set and Nellore data, explaining the precision gain without sacrificing recall. Together, these results demonstrate that selective caller intersection filtering substantially improves SV precision in low depth Illumina data while preserving recall, and that a filtering strategy optimized on human benchmarks is at least partially transferable to cattle. This provides a practical, data driven solution for reliable SV discovery in large scale livestock and non model genome studies where long read sequencing is not yet feasible.

Keywords: 2026

How to Cite:

Araújo, R., da Silva, V., Regitano, L. & Coutinho, L., (2026) “Improving structural variant discovery in low-depth Illumina sequencing using caller intersections”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286976. doi: https://doi.org/10.31274/wcgalp.24185

Rights: 1