Long-read DNA sequencing analysis: phased structural variants and SNV/indel calls from PacBio HiFi, Oxford Nanopore, or MGI CycloneSEQ reads

At a glance

What is long-read DNA sequencing?

Long-read DNA sequencing aligns or assembles individual molecules that span kilobases to megabases. That makes it possible to detect structural variants (SVs), repeat expansions, and haplotype-resolved small variants that short-read alignments often fragment across repeats [2, 7]. Pepkio maps PacBio HiFi, Oxford Nanopore, or MGI CycloneSEQ reads (when scoped) with pbmm2 or minimap2, calls SNVs/indels with Clair3 using platform-specific models and SVs with Sniffles2, and can phase variants with HiPhase or WhatsHap [1, 6, 8]. PacBio HiFi reads have a reported median per-read accuracy of 99.9% at typical CCS pass counts [5]. Custom deliverables beyond the standard workflow are scoped at kickoff.

When should you use long-read DNA sequencing?

Long-read DNA sequencing fits when the question needs SV resolution, phasing, repeat-aware variant calling, or gap closure in complex regions.

• Gapless human reference: Nurk et al. [9] assembled a complete 3.055 Gbp human genome (T2T-CHM13) using PacBio HiFi and ONT ultra-long reads, adding nearly 200 Mbp of sequence absent from GRCh38.
• Population structural variation: Reis et al. [10] profiled SVs across four Indigenous Australian communities with ONT long reads and T2T-CHM13, reporting structural variants that short-read catalogs did not capture.
• Rare-disease diagnostic variants: Höps et al. (2025) tested 100 HiFi genomes at ~30× against 145 challenging pathogenic variants; automated callers detected 83% (120/145), with 93% total after visual review.

How the analysis works — step by step

1. 1. Validate inputs and sample metadata

   Pepkio confirms FASTQ or BAM integrity (MD5 checksums), platform, chemistry, and experimental design, recording coverage, trio relationships, and reference build in sample_manifest.csv. Sub-threshold yield or missing read groups are flagged before processing [1].

   Tools and outputs

   Tools used: md5sum; samtools quickcheck; custom validation scripts

   Output: sample_manifest.csv with sample IDs, platform, read counts, batch, and QC flags

2. 2. QC raw long reads

   Read-length N50, Q-score distributions, and pass-filter rates are computed per sample; truncated libraries or low full-pass yield are flagged before alignment [1, 5].

   Tools and outputs

   Tools used: NanoPlot 1.42.0; pycoQC 2.5.2 (ONT); PacBio dataset reports (HiFi)

   Output: read_qc_summary.csv; read-length histograms; per-sample QC flags

3. 3. Generate or verify HiFi/basecalled reads

   When clients deliver raw signal (PacBio subreads, ONT POD5/FAST5, or CycloneSEQ), Pepkio runs CCS/HiFi generation, Dorado, or MGI-compatible basecalling with chemistry-appropriate models [5, 11]. Pre-delivered FASTQs are validated before filtering.

   Tools and outputs

   Tools used: PacBio SMRT Link 13.1 (pbccs / dataset); Dorado 1.4.0 (ONT raw signal when scoped); MGI CycloneSEQ basecaller (when scoped)

   Output: Platform-normalized .fastq.gz per sample; basecall_stats.csv

4. 4. Filter and normalize read sets

   Low-quality and sub-length reads are removed with documented, project-specific thresholds (Chopper for ONT/CycloneSEQ; length filters for HiFi when needed). Read loss is reported before alignment.

   Tools and outputs

   Tools used: Chopper 0.9.0; Filtlong 0.2.1 (when scoped)

   Output: Filtered .fastq.gz; filter_summary.csv with reads retained vs. removed

5. 5. Align reads to the reference genome

   HiFi reads align with pbmm2 (--preset CCS); ONT and CycloneSEQ reads align with minimap2 (-x map-ont or -x map-hifi as appropriate; [7]). Mapping rate and soft-clip profiles are audited per sample.

   Tools and outputs

   Tools used: pbmm2 1.13.1; minimap2 2.28; samtools 1.21 sort/index

   Output: Coordinate-sorted, indexed {sample}.bam and .bai; alignment_summary.csv

6. 6. Assess coverage and mappability

   mosdepth reports mean coverage and fraction of genome at ≥15× and ≥30×; samples below agreed thresholds are flagged before variant calling [1, 3].

   Tools and outputs

   Tools used: mosdepth 0.3.3; samtools 1.21 stats

   Output: coverage_summary.csv; mosdepth.global.dist.txt; coverage histogram and CDF plots

7. 7. Call small variants (SNVs and indels)

   Clair3 calls germline SNVs and indels with platform-specific models (hifi, ont; [8]); DeepVariant 1.8.0 is optional for Google pipeline parity [12]. Ti/Tv ratios and variant counts are checked per species and build.

   Tools and outputs

   Tools used: Clair3 1.0.9; DeepVariant 1.8.0 (optional)

   Output: {sample}.small_variants.vcf.gz and .tbi; per-sample SNV/indel count summary

8. 8. Call structural variants

   Sniffles2 detects DEL, DUP, INV, INS, and BND with --tandem-repeats annotations [6]. pbsv 2.9.0 is used for PacBio-native workflows when scoped [13]; cohort projects use Sniffles2 .snf merge.

   Tools and outputs

   Tools used: Sniffles2 2.8.0; pbsv 2.9.0 (PacBio-native, when scoped)

   Output: {sample}.sv.vcf.gz and .tbi; {sample}.snf for population merge; sv_count_by_type.csv

9. 9. Phase variants when metadata supports

   When trio, Hi-C, or Strand-seq phase priors are available, Pepkio phases variants with HiPhase (PacBio) or WhatsHap (long-read BAMs and phase-input VCFs; [1]). Phase block N50 is reported when phasing is in scope.

   Tools and outputs

   Tools used: HiPhase 1.6.0; WhatsHap 2.3

   Output: Phased .vcf.gz with PS/HP tags; phasing_summary.csv; phase-block length distribution plot

10. 10. Annotate, visualize, and package deliverables

    Ensembl VEP annotates small-variant consequences; SVs are overlapped with gene models and repeat annotations [14, 15]. MultiQC aggregates QC metrics. Scripts, README, Methods draft, and HTML QC report are packaged per the agreed retention policy.

    Tools and outputs

    Tools used: Ensembl VEP 112; bcftools 1.21; MultiQC 1.25; R 4.4.x / Python 3.12 plotting scripts

    Output: variant_annotation_master.tsv; sv_annotation_master.tsv; MultiQC report; final deliverable bundle

What Pepkio delivers

Technical decisions we make — and why

Reference: GRCh38 default; T2T-CHM13 when complex regions are the primary target

GRCh38 supports standard clinical annotation and cross-cohort comparison [1]. T2T-CHM13 can improve SV and alignment confidence in centromeres, telomeres, and acrocentric short arms where GRCh38 contains gaps [9, 14].

SV caller: Sniffles2 default; pbsv when scoped for PacBio

Sniffles2 is 11.8× faster and 29% more accurate than prior long-read SV callers across 5-50× HiFi and ONT data [6]. pbsv integrates with PacBio SMRT Link when clients require PacBio-native SV signatures [16].

Small-variant caller: Clair3 default; DeepVariant optional

Clair3 achieves competitive long-read SNV/indel accuracy with lower compute than graph-based short-read callers [8]. DeepVariant 1.8.0 is optional for Google benchmark parity [12].

Aligner: pbmm2 for HiFi; minimap2 for ONT and CycloneSEQ

pbmm2 wraps minimap2 with PacBio-native presets [13]. minimap2 2.28 with -x map-ont is standard for nanopore-class WGS, including ONT and CycloneSEQ FASTQs [1, 7]. CycloneSEQ tooling is newer than ONT R10.4.x; Pepkio confirms basecalling models and QC gates at kickoff [11].

Coverage gate: ≥15× for SV discovery; ≥30× for joint small-variant + SV clinical interpretation

In simulated benchmarks, multiple SV callers exceeded F1 0.75 at ~20× [3]. Höps et al. (2025) validated HiFi diagnostic panels at ~30×; 90% of automatically called variants remained detectable at 15× in titration analysis.

Common questions

Related services

• Whole-genome sequencing — Lower-cost short-read SNV/indel catalogs when long-read SV resolution is not needed.
• CNV and structural variation — Short-read CNV/SV calling when long-read data are unavailable.
• Variant calling — Caller selection, filter tuning, and joint genotyping when alignment is already complete.
• Long-read RNA-seq — Isoform-resolved transcriptome analysis that complements DNA SV and phasing context.
• Custom consulting — Pre-sequencing depth, platform, and reference-build planning before library prep.