Introduction
Accurate haplotype phasing is critical for improving our understanding of how genetics contributes to human health and disease.1 For example, compound heterozygotes have two different pathogenic variants within the same gene. Some genes are associated with autosomal recessive conditions, including disorders of sex development, congenital heart disease, and cancer.2 In these genes, having two pathogenic variants on opposite copies of the gene (in trans) may be sufficient for phenotypic expression. Furthermore, haplotyping is critical in the determination of allele-specific expression, which may affect human health (e.g., certain haplotypes in CHGA affect the risk of hypertensive renal disease).3 In addition, the generation of haplotype-resolved assemblies improves detection of structural variants and interrogation of highly polymorphic medically important regions of the genome, facilitating precision medicine.4
While homozygous loci do not differ and thus the parental origin of each allele cannot be determined, the presence of alternate alleles allows heterozygous loci to be resolved relative to one another (Figure 1A). The group of genetic variants inherited as a unit from a single parental chromosome is known as a haplotype. Alternate alleles that are inherited together are in a cis configuration, while alternate alleles that map to separate chromosomes are in trans.
Read-based phasing can be used to deduce haplotypes by analyzing sequencing reads that span more than one heterozygous variant. A phase block is an assembled chain of sequencing reads in which heterozygous variants are confidently phased together (Figure 1B).5 Ideally, phasing would produce one phase block per chromosome, with all heterozygous variants resolved relative to each other. In practice, however, phase blocks cannot be joined when sequencing data is absent or contradictory. Furthermore, even within genes that reside on a single phase block, not all heterozygous variants are phased. Some variants lack coverage while others are excluded based on variant type or call quality. Collectively, variant phasing quality can be affected by factors such as phasing errors, phase block length, and contiguity, as described in Table 1.
Figure 1. Depiction of haplotype phasing and phase blocks
A) Haplotype phasing reveals which variants are located together on the same chromosome. In this example, the middle alternate allele on haplotype 1 is in trans to the flanking alternate alleles on haplotype 2, while the flanking haplotype 2 alleles are in cis to one another. B) A phase block is a chain of co-phased heterozygous variants that a phasing algorithm was able to confidently resolve from sequencing reads.
The proximity-mapped read technology of TruPath™ Genome6,7 provides a novel approach to read-based phasing using short-read sequencing. Reads from the same DNA fragment are grouped together due to their proximity, which connects heterozygous variants even when no individual read spans both loci. This proximal linking creates more opportunities for heterozygous sites to be sampled together, thereby increasing the likelihood of co-phasing. We evaluated the performance of TruPath Genome compared to publicly released vendor data from long-read platforms Pacific Biosciences (PacBio) and Oxford Nanopore (ONT) across a comprehensive set of phasing quality metrics.
Phasing metrics
Comprehensive evaluation of phasing quality is multi-dimensional (Figure 2). The four dimensions of phasing quality considered here are:
- Completeness—percentage of heterozygous variants assigned to a haplotype
- Contiguity—length of uninterrupted phase blocks
- Genic coverage—fraction of genes fully and correctly phased
- Accuracy—rate of haplotype assignment errors
Figure 2. Four dimensions of phasing quality
Contiguity, accuracy, and genic dimensions can be assessed using multiple measures (Table 1). Measures can either require a truthset or be agnostic of one (e.g., NIST Genome in a Bottle HG002 v5.0q truthset). Each dimension is an independent measurement. A platform can excel at one metric while failing at another. Because an honest assessment of phasing quality should report across all four dimensions, our analyses included metrics from each of the dimensions.
Table 1. Phasing quality metrics
To cut through the complexity of traditional phasing benchmarks, we prioritized a streamlined set of four core metrics (highlighted below) that cleanly capture completeness, physical span (via D80 over NG50), and functional accuracy at the gene and variant level.
Comparative phasing performance
Challenges to direct cross-platform comparison include differing read length distributions, error profiles, and downstream phasing pipelines. For this comparison, only one sample, HG002, had publicly released data available for Illumina (ICA, BaseSpace), PacBio (FTP link), and ONT (FTP link). Of note, data inputs were not depth-normalized. Each platform was assessed using the vendor-recommended configuration; TruPath Genome results were generated using DRAGEN 4.5. Only small variant phasing, comprising single nucleotide polymorphisms (SNPs) and indels, was considered. When possible, phasing performance was assessed using WhatsHap5, supplemented by an internally developed tool that extends existing WhatsHap metrics and introduces novel metrics.
Completeness, accuracy, and percentage of correctly phased heterozygous variants
Completeness, measured as the percentage of heterozygous variants phased, was highest for TruPath Genome (Figure 3). The overall percentages of heterozygous variants phased were similar for TruPath Genome and PacBio, at 99.0% and 98.1%, respectively, while ONT had 75.2% of variants phased. Although ONT had a high percentage of phased heterozygous SNPs, it did not phase indels.
Figure 3. Measures of completeness across three platforms
TruPath Genome and PacBio have similarly high percentages of heterozygous variants phased.
Contiguity
Despite using short-read sequencing, TruPath Genome produces the longest phase blocks in this analysis with a median NG50 of 7.0 Mb, compared to 0.8 Mb and 1.3 Mb for PacBio and ONT, respectively (Figure 4A). Consistent with this, the median TruPath D80 is 1.3 Mb, compared with 228 kb for PacBio. D80 cannot be calculated for ONT when both SNPs and indels are considered because unphased indels keep phasing yield below 80% at every measured distance (Figure 4B). When the analysis is limited to SNPs, D80s are 1.3 Mb, 231 kb, and 274 kb for TruPath Genome, PacBio, and ONT, respectively (Figure 4C).
Figure 4. TruPath Genome demonstrates improved phasing contiguity
TruPath Genome produces longer phase blocks than either long-read platform (Figure 4A). Its higher D80 indicates that heterozygous variants remain co-phased across greater genomic distances when all variants are considered (Figure 4B). TruPath Genome maintains its advantage when the analysis is limited to SNPs (Figure 4C).
Figure 5. Phasing yield curves for one sample per technology, calculated across all variants and SNPs only.
In the datasets evaluated, TruPath Genome consistently phases heterozygous variants to each other over longer ranges than those achieved with long-read platforms.
Accuracy
Error rates are an important measurement in determining the accuracy of phasing. Errors include switch, flip, and switch-flip errors (Figure 6A). Within the comparison among TruPath Genome, PacBio, and ONT, median switch/flip and non-flip switch rates were <0.1% across all platforms (Figures 6B and 6C).
Figure 6. Phasing error rates
Switch, flip, and switch-flip error rates measure how often phase changes between adjacent sites (Figure 6A). Low (<0.1%) switch/flip (Figure 6B) and non-flip switch error (Figure 6C) rates suggest that the superior phasing completeness of TruPath Genome does not come at the expense of accuracy.
Genic
As seen in Figure 7A, almost all (98.1%) genes are fully phased using TruPath Genome, higher than both PacBio (83.9%) and ONT (87.9%). When considering the percentage of heterozygous pairs co-phased within a gene (Figure 7B) and the percentage of these pairs that were correctly phased (Figure 7C), TruPath Genome had higher values (97.7% and 97.6%) compared to PacBio (81.7% and 82.3%) and ONT (63.3% and 67.6%).
Figure 7. Evaluation of genic measurements
In the datasets evaluated, TruPath Genome had the highest fraction of fully phased genes (Figure 7A), the highest percentage of heterozygous pairs co-phased within a gene (Figure 7B), and the highest percentage of pairs with the correct phase relationship (Figure 7C). Note that percent pairs correct is calculated over all heterozygous variants within a gene and will be impacted by both phasing accuracy and phasing contiguity.
TruPath Genome performance with standard DNA inputs
It is important to note that all analyses presented above used high molecular weight (HMW) DNA. HMW DNA extraction is more laborious and costly than standard DNA isolation and may not be available to all laboratories. Because TruPath Genome relies on proximity read mapping, phasing can also be successfully performed using standard molecular weight DNA. Standard extractions with TruPath provide similar completeness and contiguity to traditional long-read sequencing technologies in the datasets evaluated here (Figures 8A, 8B, and 8C). Although error rates were higher when using standard molecular weight DNA, switch/flip and non-flip switch error rates were <0.5% for both (Figures 9A and 9B).
Figure 8. Comparison of phasing completeness metrics between HMW and standard molecular weight DNA when using TruPath Genome
Figure 9. Comparison of phasing accuracy metrics between HMW and standard molecular weight DNA when using TruPath Genome
Importantly, genic metrics from TruPath analyses of standard DNA extractions, although lower than those generated with HMW DNA (Figure 10), are higher than those observed with traditional long-read sequencing methods.
Figure 10. Comparison of genic phasing metrics between HMW and standard molecular weight DNA when using TruPath Genome
Conclusion and next steps
The launch of TruPath Genome in early 2026 allows users to perform long-range phasing using short-read sequencing. Using metrics that span all four dimensions of phasing quality, we demonstrate that TruPath Genome achieves greater completeness, contiguity, and genic coverage than commercially available long-range sequencing platforms while maintaining similarly high accuracy in this analysis.
TruPath Genome is a new assay with novel informatics that evolves as the assay matures. Future DRAGEN™ software releases are intended to improve phasing performance across all four metric dimensions and expand the range of variant types supported for phasing.
For Research Use Only. Not for use in diagnostic procedures.
References
- Tewhey R, Bansal V, Torkamani A, Topol EJ, Schork NJ. The importance of phase information for human genomics. Nat Rev Genet. Mar 2011;12(3):215-23. doi:10.1038/nrg2950
- Miller DB, Piccolo SR. A Survey of Compound Heterozygous Variants in Pediatric Cancers and Structural Birth Defects. Original Research. Frontiers in Genetics. 2021-March-22 2021;Volume 12 - 2021. doi:10.3389/fgene.2021.640242
- Chen Y, Rao F, Wen G, et al. Naturally occurring genetic variants in human chromogranin A (CHGA) associated with hypertension as well as hypertensive renal disease. Cell Mol Neurobiol. Nov 2010;30(8):1395-400. doi:10.1007/s10571-010-9600-2
- Garg S, Fungtammasan A, Carroll A, et al. Chromosome-scale, haplotype-resolved assembly of human genomes. Nat Biotechnol. Mar 2021;39(3):309-312. doi:10.1038/s41587-020-0711-0
- Martin M, Patterson M, Garg S, et al. WhatsHap: fast and accurate read-based phasing. openRxiv; 2016.
- Illumina. Introducing constellation mapped read technology. Genomics Research Hub blog. 2024. https://www.illumina.com/science/genomics-research/articles/constellation-mapped-read-technology.html
- Illumina. Informatics advances reveal the TruPath™ Genome towards comprehensive genomic insights. GENOMICS RESEARCH HUB blog. 2026. https://www.illumina.com/science/genomics-research/articles/informatics-advances-reveal-the-trupath-genome-towards-comprehen.html