PacBio HiFi Variant Calling: Accuracy And Workflow Choices

PacBio HiFi (High-Fidelity) long reads provide single-molecule accuracy above 99.9% and read lengths averaging 10,25 kb, making them the current gold standard for detecting single-nucleotide variants (SNVs), small indels, and structural variants (SVs) in complex genomic regions. This guide is for bioinformaticians, clinical geneticists, and molecular biologists who need a source-bounded, practical framework for choosing tools, running pipelines, and interpreting results. You will leave with clear decision criteria, a tested workflow, and awareness of common pitfalls. For authoritative background on sequencing technology, refer to the NCBI Bookshelf entry on long-read sequencing.

At a Glance

The key takeaway: HiFi variant calling is not simply “short reads with longer reads.” You must adjust coverage, mapping parameters, and caller settings to exploit long-read synergy. Use the Galaxy Training Network for hands‑on tutorials that walk through each step.

Core Concepts and Accuracy

PacBio HiFi reads are generated by repeatedly sequencing a circular DNA template. The consensus sequence from the multiple passes (typically 10,30 passes per read) yields a per‑base error rate below 0.1%. This accuracy is uniform across GC‑rich and GC‑poor regions, unlike short‑read technologies that struggle with extreme GC content. For a deep dive into accuracy benchmarks, read HiFi sequencing accurately identifies clinically relevant variants in paralogous genes (Am J Hum Genet). That study showed that HiFi outperformed short reads in discriminating variants in the PRORP and PMS2 paralogous regions, where short reads routinely produce false positives.

Why does accuracy matter for variant calling? Higher read accuracy reduces the number of spurious variant calls that arise from sequencing errors. When a caller like DeepVariant processes HiFi reads, it can use a smaller training set and still achieve recall rates above 99.5% for SNVs. The tradeoff is that HiFi coverage must be sufficiently deep to overcome residual systematic errors in homopolymer runs (e.g., AAAA...). The EMBL‑EBI Training materials on long‑read analysis provide detailed error profiles.

Decision Points for Workflow

You must decide on three factors before running any pipeline.

1. Read alignment strategy. Minimap2 is the de facto standard for HiFi reads. It uses the map‑ont preset (‑x map‑hifi) to optimize for low error rates and long alignments. Some groups use pbmm2 (a wrapper of minimap2 from PacBio) for easier integration. If your reads have strong methylation signals (e.g., from HiFi kinetic data), you may want to consider an aligner that retains BAM tags for downstream methylation analysis. See Genome‑wide methylation detection and episignature analysis using PacBio long‑read sequencing (Genome Med) for that specialized case.

2. Variant caller selection. For high‑confidence SNVs and small indels, DeepVariant with the PacBio HiFi model is the current best-in‑class. For structural variants, you have three popular options: pbsv (PacBio’s own caller), Sniffles2, or cuteHap. cuteHap: Haplotype‑Aware Structural Variant Detection in Phased Long‑Read Sequencing Data (Adv Sci) describes a tool that leverages phased reads to resolve SVs in diploid genomes. If you work on cancer or somatic calling, consider using a caller that can handle variable allele frequencies, such as Long‑Shot for SNVs.

3. Phasing requirements. HiFi reads are long enough to span multiple heterozygous variants, allowing haplotype phasing without parental data. Tools like Whatshap or the phasing module included in DeepVariant’s pipeline can produce phased VCFs. Phased data is critical for detecting compound heterozygosity and for interpreting diagnostic utility of clinical genome reanalysis in rare pediatric disorders using long‑read sequencing (HGG Adv). If you need phase, include a phasing step in your workflow.

Practical Workflow

Here is a reproducible step‑by‑step pipeline, assuming you have HiFi reads in FASTQ format. All tools are available through Bioconda or container images.

Step 1: Quality control and read trimming Run FastQC on the raw reads. Then use HiFiAdapterFilt to remove adapter sequences. A small percentage of HiFi reads may contain chimeric artifacts from incomplete circular concatemers. Filter those out with a custom script that checks for read length drops. Store the cleaned reads. The NCBI Sequence Read Archive holds public HiFi datasets you can use for testing.

Step 2: Alignment Use minimap2 with the preset ‑x map‑hifi. Example command:

minimap2 -ax map-hifi reference.fa cleaned_reads.fastq > aligned.sam

Convert to sorted BAM with samtools sort and index. Evaluate mapping rate (should be >95%) and median insert size. For genome assemblies, use the Haplotype‑resolved genome assemblies of BJ and IMR‑90 human fibroblast cell lines (Nucleic Acids Res) as a reference for alignment quality metrics.

Step 3: SNV and indel calling Run DeepVariant with the --model_type PACBIO flag. Provide the BAM and reference FASTA. DeepVariant outputs a VCF and a gVCF. For cohort studies, merge gVCFs with glnexus and then joint‑call. Validate known variants using a truth set (e.g., GIAB HG002).

Step 4: Structural variant calling Call SVs with pbsv or cuteHap. Example for pbsv:

pbsv discover aligned.bam | pbsv call reference.fa output.vcf

Filter SVs that have low support (e.g., <3 reads) or are in low‑mappability regions. Because HiFi reads can span 20 kb, you will detect large insertions and deletions that short reads miss. Pay special attention to SVs in clinically relevant regions such as the 2q31.1 locus described in De Novo Complex Genomic Rearrangement Spanning 2q31.1 in a Proband With Congenital Malformations (Am J Med Genet A).

Step 5: Phasing (optional) If phased output is required, use whatshap phase on the SNV VCF. This tool uses read‑backed phasing from the long HiFi reads. The phased VCF can be used to resolve compound heterozygous candidates.

Quality Checks

After calling, perform these checks:

• Coverage distribution. Plot depth per chromosome. HiFi coverage can be uneven if there are biases in library prep. Regions with coverage below 10x should be flagged as low‑confidence.
• Transition / transversion ratio (Ti/Tv). For human genome WGS, Ti/Tv should be about 2.0,2.1. Lower ratios suggest false positive calls.
• Indel ratio. The number of heterozygous to homozygous indels should be around 1:2 for autosomal calls. Extreme skew indicates mapping artifacts.
• SV call consistency. Overlap your SV calls with a tandem repeat annotation. SVs that exactly span known tandem repeats are often artifactual if the reads have noise in repeat counts.
• Comparison with orthogonal data. If available, compare SNV calls with short‑read calls from the same sample. Concordance should be >99.5% for high‑confidence regions. Discrepancies often arise in segmental duplications and low‑complexity regions (exactly where HiFi excels).

Common Mistakes

1. Under‑sequencing coverage. Many users assume that because HiFi reads are accurate, 10x coverage is enough. In practice, 15x is the minimum for SNVs, and 30x is recommended for structural variants. Below these thresholds, false‑negative rates increase sharply.

2. Using short‑read tools without adjustment. Tools like GATK HaplotypeCaller are not designed for long reads. They may produce errors in region with large indels. Always use HiFi‑trained models (DeepVariant’s PACBIO model) or long‑read‑aware callers.

3. Ignoring read‑level artifact flags. HiFi reads occasionally contain chimera from incomplete circularization. If a variant is supported only by reads with adapter contamination, it is likely a false positive. Filter those reads out before calling.

4. Over‑interpreting SV breakpoint precision. Although HiFi reads define breakpoints more accurately than long noisy reads, systematic errors remain near homopolymers and simple repeats. The exact base at a breakpoint may be off by 1,2 bases. When reporting SVs, give a confidence interval.

5. Not splitting by haplotype. For diploid genomes, calling SVs without phasing can miss allelic deletions or duplications. A deletion that appears homozygous may actually be hemizygous if the reads from one haplotype are missing due to alignment failure. Use phased calls to disambiguate.

Limits of Interpretation

PacBio HiFi variant calling is powerful but not infallible. Recognize these boundaries:

• Homopolymer and simple repeat regions remain problematic because the circular consensus algorithm can fractionally underestimate or overestimate the number of identical bases. Variants in such regions should be validated with Sanger sequencing or a complementary method.
• Copy‑ number variants (CNVs) are less reliably called with standard SV callers. HiFi reads alone do not provide the depth‑of‑coverage signal needed for accurate copy‑number estimation. Use a dedicated tool like HiFi‑CNV or combine with short‑read data.
• Methylation detection is possible because kinetically modified bases affect polymerase speed, but base‑level methylation calls are not as precise as bisulfite sequencing. The Genome‑wide methylation detection and episignature analysis paper demonstrates that HiFi can capture genome‑wide methylation patterns but at lower resolution than dedicated methods.
• Somatic variant calling in low‑purity tumor samples (allele fraction <10%) is difficult because HiFi coverage is typically low (20,30x). For somatic SNVs with VAF under 5%, use a caller that models sequencing error specifically (e.g., Strelka2 in long‑read mode, but it is not fully optimized).
• Contamination and chimeric reads can inflate variant calls. Always run a contamination check (e.g., verifyBamID) and remove reads that map to multiple chromosomes with identical high scores.

Frequently Asked Questions

1. What coverage is recommended for detecting de novo mutations with PacBio HiFi? For de novo SNV detection with high specificity, aim for 30x coverage per parent and proband. At that depth, most trinucleotide repeat expansions will be captured. At lower coverage, you risk missing or mis‑calling de novo events. Verify with orthogonal short‑read data in candidate regions.

2. Can I use PacBio HiFi data for HLA typing? Yes. HiFi reads span the entire HLA class I and II genes (4,5 kb each), so they provide unambiguous phased alleles. The accuracy is superior to short‑read imputation. Use tools like HLA‑PRG or OptiType with long‑read support.

3. How do I compare variant calls from HiFi with calls from Illumina? Align both VCFs to the same reference. Focus on high‑confidence intervals (e.g., GIAB v4.1 for human). Compute precision and recall per variant type. Discrepancies in repetitive regions often favor HiFi calls. Report concordance metrics as a table in your output.

4. Is it necessary to run a separate caller for structural variants? Yes. SNV callers like DeepVariant do not output SVs. You need a dedicated tool (pbsv, Sniffles2, or cuteHap). However, some pipelines (e.g., PEPPER‑Margin‑DeepVariant) now integrate SV detection. Check the documentation of your chosen pipeline.

References and Further Reading

• NCBI Bookshelf: An Introduction to Long‑Read Sequencing , Foundational technical overview.
• Galaxy Training Network: PacBio HiFi Data Analysis , Step‑by‑step tutorials for alignment and calling.
• EMBL‑EBI Training: Long‑Read RNA‑Seq and Variant Calling , Free courses covering quality control and phasing.
• Bioconductor: Analysis of PacBio HiFi Methylation Data , R packages for downstream analysis of epigenetic marks.
• HiFi sequencing accurately identifies clinically relevant variants in paralogous genes (Am J Hum Genet) , Key benchmark against short reads.
• Diagnostic utility of clinical genome reanalysis using long‑read sequencing (HGG Adv) , Highlights reanalysis case studies.
• Haplotype‑resolved genome assemblies reveal extensive structural variation (Nucleic Acids Res) , Practical insights for SV interpretation.
• De Novo Complex Genomic Rearrangement Spanning 2q31.1 (Am J Med Genet A) , Example of HiFi resolving complex rearrangements.
• cuteHap: Haplotype‑Aware Structural Variant Detection (Adv Sci) , Novel method for phased SVs.
• Genome‑wide methylation detection and episignature analysis using PacBio long‑read sequencing (Genome Med) , Epigenetic application of HiFi.

Related Articles

• Protein Synthesis
• Incomplete Dominance Gene
• Cell Membrane Function Biology
• Dna Structure
• Protein Structure