Systematic benchmarking of tools for structural variation detection using short- and long-read sequencing data in pigs.

Evaluating diverse structural variation (SV) detection-relevant programs leveraging different algorithms has become a pressing need in humans and farm animals. We addressed this by sequencing five genetically diverse pig individuals (breeds) with short- and long-read DNA-sequencing platforms. We created the SV benchmark set for each breed and assessed the performance of 16 SV calling-relevant tools. Results showed that long-read platforms enabled detecting many SVs missed by short-read platforms with similar precision. Benchmark SVs, mainly 200-500 bp insertions/deletions, had high validation rates. The assembly-based SV calling program SVIM-asm showed superior detection performance and resource consumption. The SVs with more supporting reads, sizes under 1 kb, outside simple repeat area, in low GC content and runs of homozygosity regions, had higher detection accuracy. Alignment-based tools performed well even at 5 depth. Our study provides systematic guidance for an optimal SV calling pipeline in pigs and other farm animals.