Improving long-read somatic structural variant calling with pangenome and de novo personal genome assembly

mminisv filtered structural-variant callsets Overview This Zenodo dataset contains structural-variant (SV) callsets processed with the mminisv filtering workflow. The original SV calls were generated by four SV callers: SAVANA Severus Sniffles2 nanomonsv Two alternative filtering strategies are provided for each available sample, sequencing platform, and caller combination: l+s: filtered using evidence from de novo assembly. l+g: filtered using evidence from the HPRC v2 graph genome. The record also includes the COLO829 SV ground-truth set for benchmarking. Dataset summary Item Count Samples 6 Sample-platform combinations 7 SV callers 4 Filtering strategies 2 l+s filtered callsets 28 l+g filtered callsets 28 Ground-truth callsets 1 Total VCF files 57 Samples and sequencing platforms Sample HiFi (hifi1) ONT (ont1) Number of filtered VCFs COLO829 Yes Yes 16 HCC1395 Yes No 8 HCC1937 Yes No 8 HCC1954 Yes No 8 NCI1437 Yes No 8 NCI2009 Yes No 8 Each sample-platform combination contains callsets from all four callers under both filtering strategies. File naming convention Filtered callsets follow this pattern: ensembl__<sample>_<platform>__<caller>_<filter>_3_filtered.vcf Example: ensembl__COLO829_hifi1__savana_l+s_3_filtered.vcf Filename fields are interpreted as follows: Field Values or meaning <sample> COLO829, HCC1395, HCC1937, HCC1954, NCI1437, or NCI2009 <platform> hifi1 or ont1 <caller> savana, severus, sniffles2, or nanomonsv <filter> l+s for de novo assembly filtering or l+g for HPRC v2 graph-genome filtering _3_filtered Pipeline-specific suffix retained in this data release The + characters in l+s and l+g are literal parts of the filenames. Quoting filenames in shell commands is recommended. Ground-truth callset COLO829.truth.hs38.vcf This file contains the COLO829 SV ground-truth set and is intended for benchmarking the COLO829 callsets. Confirm reference-build and contig-name compatibility before comparison. Data organization The 56 filtered callsets form the following complete combination matrix: 7 sample-platform combinations × 4 SV callers × 2 filtering strategies = 56 filtered VCF files Sample-platform combinations: COLO829_hifi1 COLO829_ont1 HCC1395_hifi1 HCC1937_hifi1 HCC1954_hifi1 NCI1437_hifi1 NCI2009_hifi1 Callers: savana severus sniffles2 nanomonsv Filtering strategies: l+s l+g Suggested applications These data can be used to: benchmark filtered COLO829 calls against the supplied SV truth set; compare de novo assembly filtering (l+s) with graph-genome filtering (l+g); compare caller-specific SV outputs; compare HiFi and ONT results for COLO829; and evaluate the consistency of SV calls across the included samples. Basic validation Count the filtered callsets after downloading: find . -maxdepth 1 -name 'ensembl__*_l+s_3_filtered.vcf' | wc -l # Expected: 28 find . -maxdepth 1 -name 'ensembl__*_l+g_3_filtered.vcf' | wc -l # Expected: 28 Check that all VCF headers can be read by bcftools: for vcf in ./*.vcf; do if ! bcftools view -h "$vcf" >/dev/null; then echo "Invalid or unreadable VCF: $vcf" >&2 fi done Inspect a callset: bcftools view -h \ 'ensembl__COLO829_hifi1__savana_l+s_3_filtered.vcf' bcftools view -H \ 'ensembl__COLO829_hifi1__savana_l+s_3_filtered.vcf' | head Verify a downloaded file against the MD5 checksum reported by Zenodo: md5sum 'ensembl__COLO829_hifi1__savana_l+s_3_filtered.vcf' Benchmarking considerations SV callers may encode equivalent events differently, particularly translocations, breakends, insertions, and complex rearrangements. Before record-level comparison, harmonize reference builds, contig names, SV representations, size thresholds, and breakpoint-matching criteria. The l+s and l+g files are alternative filtered versions of related underlying caller outputs. They should not be treated as independent biological replicates. Only COLO829 has both HiFi and ONT callsets in this release. Cross-platform comparisons are therefore limited to COLO829. Access and citation Zenodo record: 21852833 Please cite the Zenodo record and the corresponding mminisv workflow or publication when using these data. License and formal citation metadata should be taken from the published Zenodo record.