Back

The Genome In A Bottle HG002 assembly-based variant benchmark set enables comprehensive benchmarking of small and structural variants.

Olson, N. D.; Dwarshuis, N.; Hansen, N. F.; Koren, S.; Rhie, A.; McDaniel, J. H.; Wagner, J.; Berry, G.; Shafin, K.; English, A.; Saunders, C. T.; Kundu, R.; Holt, J. M.; Freed, D.; Abdulkadir, A. A.; Daniels, C.; Majidian, S.; Paulin, L. F.; Sedlazeck, F. J.; Fleharty, M.; Gao, Y.; Denti, L.; Brambrink, L.; Carroll, A.; Chang, P.-C.; Cook, D. E.; Kolesnikov, A.; Catreux, S.; Edlund, C.; Han, J.; Murray, L.; Qiu, Y.; Rossi, M.; Scheffler, K.; Truong, S.; Wang, Y.; Chikhi, R.; Flores, C.; Jaspez, D.; Lorenzo-Salazar, J. M.; Munoz-Barrera, A.; Rubio-Rodriguez, L. A.; Lin, M.-J.; Vaddadi, K.; Lan

2026-10-01 genomics
10.64898/2026.09.23.752440 bioRxiv
Show abstract

High-accuracy, high-resolution benchmark data are fundamental to improving the accuracy and generalizability of sequencing technologies and associated algorithms. Most existing benchmarks for DNA sequencing rely on mapping to a single linear reference, omitting inaccessible and highly repetitive complex genomic regions. Here, we present the Genome in a Bottle (GIAB) Consortium's HG002 v5.0q benchmark set, a more comprehensive variant benchmark derived from a fully phased, highly curated, and accurate diploid assembly of HG002. Our approach leverages recent advances in long-read sequencing and assembly to provide confident small- and structural-variant calls across previously inaccessible regions, including segmental duplications, homopolymers, tandem repeats, and complex immune loci. The v5.0 benchmark significantly expands the callable genome relative to v4.2.1, adding x200 Mbp of sequence and increasing variant detection by 18 % for small variants and 3-fold for structural variants compared to previous benchmarks. The benchmark regions include more variants in repetitive regions, including 1.2 x segmental duplications, 1.8 x tandem repeats, and 2.0 x homopolymers, compared to v4.2.1. Separate and combined small- and structural-variant benchmarks are available for three versions of the human reference genome (GRCh37, GRCh38, and CHM13). We validate the reliability of this benchmark set by comparing multiple sequencing technologies and variant calling pipelines and curating differences. The v5.0q benchmark provides a critical resource for the development and validation of next-generation sequencing technologies and analysis methods.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.